From 1 September 2026, Microsoft Entra ID now treats passkeys as the default authentication method for users who still rely on SMS or voice verification.

Microsoft will automatically enable those users for passkeys as the rollout reaches their organisation.

The next time they complete multifactor authentication, Entra ID can prompt them to register a passkey.

The change marks the beginning of a much bigger transition.

On 1 February 2027, Microsoft plans to stop providing SMS and voice authentication natively through Entra ID.

For businesses that still depend on verification codes sent to phones, the clock has started.

Microsoft wants businesses to move beyond SMS codes

SMS-based multifactor authentication helped companies add a second layer of security beyond passwords.

However, attackers have become much better at defeating it.

Criminals can steal verification codes through phishing sites. They can also use social engineering, SIM-swap attacks and other methods to intercept phone-based authentication.

Passkeys work differently.

Instead of sending a secret code that a user types into a website, a passkey uses cryptographic credentials.

The private part of the credential stays on the user’s device or inside a trusted credential manager.

The website or service receives proof that the correct credential exists.

There is no reusable code for an attacker to steal.

That makes passkeys resistant to many common phishing attacks.

What changes for Entra ID users now?

Microsoft is not disabling SMS authentication immediately.

The September change starts the migration.

Users who have SMS or voice enabled can automatically become eligible for passkeys.

When they next complete an MFA sign-in, Microsoft can ask them to create one.

Users can initially postpone the registration prompt.

That gives companies time to prepare employees and update internal support processes.

However, organisations should not treat that flexibility as permanent.

Microsoft has already set the next deadline.

From 1 February 2027, the company will stop delivering SMS and voice authentication messages itself.

Businesses that fail to prepare could face sign-in problems.

What happens on 1 February 2027?

The February deadline creates the biggest change.

Microsoft-provided SMS and voice authentication will end in public-cloud Entra ID environments.

Users who only have SMS or voice available will then need another authentication method.

If they do not have one, Microsoft will require them to register a passkey before they can continue signing in.

Unlike the earlier migration prompts, companies will not be able to opt out of that requirement.

Microsoft recommends moving employees to passkeys or another phishing-resistant method before the deadline.

Businesses that genuinely need SMS or voice will still have another option.

They will need to use a third-party telecommunications provider instead of Microsoft handling those messages directly.

Businesses can still keep SMS if they really need it

Microsoft is not removing the ability to use phone-based verification entirely.

Some organisations operate in industries where regulations or unusual technical environments may still require SMS or voice.

Those businesses will be able to choose a telecommunications provider through the Microsoft Security Store.

The company plans to make information about supported providers available from 18 September 2026.

Administrators will be able to configure supported providers from 30 October 2026.

The organisation will then manage the relationship with that provider.

Microsoft clearly sees this as an exception rather than the preferred path.

For most users, it wants passkeys to replace SMS.

Entra ID supports more than one type of passkey

Microsoft supports both synced and device-bound passkeys.

A synced passkey can live inside a credential manager and move between a user’s devices.

Examples include passkeys stored through Google Password Manager or Apple’s iCloud Keychain.

This approach can make passkeys easier for ordinary employees because they follow the user across supported devices.

A device-bound passkey stays tied to a specific device or security key.

Microsoft Authenticator can store device-bound credentials.

Windows devices can also use Entra passkeys.

Businesses can use physical FIDO2 security keys for employees who need stronger hardware-backed authentication.

Organisations can therefore choose different options for different types of users.

Passkeys remove one of phishing’s most useful tricks

Passwords and verification codes share a major weakness.

Users can type them into the wrong website.

An attacker can build a convincing copy of a Microsoft sign-in page and persuade an employee to enter a password.

The attacker can then ask for the MFA code.

If the victim supplies it quickly enough, the criminal may gain access.

A passkey checks the website it belongs to.

A credential created for Microsoft’s legitimate service will not authenticate an attacker-controlled phishing site.

That makes the sign-in process much harder to relay through a fake page.

Passkeys also remove the need for users to read and type temporary codes.

For businesses, stronger security could therefore come with a simpler login process.

This matters even more as AI improves phishing

Microsoft has linked the authentication change to the changing security environment.

Generative AI has made it easier to create convincing phishing messages.

Attackers can produce better-written emails, personalised messages and believable fake conversations at scale.

They can also automate more parts of an attack.

This increases the value of security controls that do not depend on a person spotting every scam.

Security awareness training still matters.

However, even well-trained employees make mistakes.

A phishing-resistant authentication method can block an attack even when a user clicks the wrong link.

That is one reason Microsoft is moving passkeys from an optional security feature towards the default.

South African businesses should start checking their Entra environments

The change has direct relevance to South African organisations.

Microsoft Entra ID forms part of the identity infrastructure behind many Microsoft 365 and enterprise deployments.

Businesses, schools, government organisations and other institutions use it to control access to applications and cloud services.

IT departments should identify employees who still rely on SMS or voice MFA.

They should then plan how those users will move to passkeys or another supported phishing-resistant method.

Large organisations should not wait until January 2027.

A migration can create support questions.

Employees may need help registering credentials on phones, laptops or hardware security keys.

Companies also need recovery procedures for lost or replaced devices.

Testing those processes before Microsoft removes native SMS delivery will reduce the risk of disruption.

Passkeys do not mean passwords disappear everywhere overnight

The word “passkey” often gets mixed up with the broader idea of passwordless computing.

The two concepts overlap, but businesses should not assume this change instantly removes every password.

An employee may still encounter passwords in older applications or systems that do not support modern authentication.

Some organisations also use several identity platforms.

Microsoft’s change specifically targets the authentication experience in Entra ID.

Its goal is to move more users towards phishing-resistant credentials.

The wider transition away from passwords will take longer.

IT departments can temporarily delay the September change

Microsoft does provide a temporary opt-out during the transition period.

Administrators that need extra time can prevent the automatic passkey enablement and registration campaign from applying immediately.

That option lasts only during the migration window.

It does not cancel the February 2027 deadline.

From that date, Microsoft will enforce the new sign-in requirements across affected public-cloud tenants.

Businesses that use the temporary delay should therefore treat it as preparation time rather than a way to avoid the change.

Microsoft is making identity security a bigger part of its platform

Microsoft’s decision reflects a broader change in enterprise cybersecurity.

Identity has become one of the most valuable targets for attackers.

A stolen employee account can provide access to email, documents, cloud services and business applications.

Administrative accounts create even greater risks.

Microsoft has responded by putting more security controls into Entra, Windows and its wider cloud platform.

Passkeys fit into that strategy.

The company already supports Windows Hello for Business, FIDO2 security keys and other passwordless authentication options.

Making passkeys the default pushes those technologies towards a much larger group of users.

Who is Microsoft today?

Microsoft was founded in 1975 by Bill Gates and Paul Allen.

The company built its early success around software for personal computers before Windows and Office turned it into one of the world’s largest technology businesses.

Today, Microsoft operates across cloud computing, enterprise software, cybersecurity, gaming, artificial intelligence and consumer technology.

Satya Nadella serves as chairman and chief executive officer.

Microsoft Azure has become one of the company’s most important platforms, while Microsoft 365 remains central to its enterprise business.

Entra forms Microsoft’s identity and access-management portfolio.

It includes Entra ID, the service previously known as Azure Active Directory.

The company renamed Azure AD to Microsoft Entra ID in 2023 as it expanded the Entra security brand.

The February deadline matters more than the September prompt

For employees, the visible change may begin with a simple message asking them to create a passkey.

For IT departments, the significance runs much deeper.

Microsoft has now set an end date for one of the most common enterprise MFA methods it provides.

SMS authentication will not vanish from the world on 1 February 2027.

However, Microsoft no longer wants to provide it as the default safety net for Entra customers.

Businesses now have two choices.

They can move users towards phishing-resistant authentication.

Or they can take responsibility for maintaining phone-based authentication through another provider.

Microsoft has made its preferred option clear.

The future of Entra ID sign-ins is built around passkeys, not text messages.

The post Microsoft is pushing Entra ID users away from SMS passkeys are now the default appeared first on Technology Blog.

The company announced both models on 1 September 2026.

Fable 5.1 is Anthropic’s most capable generally available model for coding, knowledge work and long-running AI agents.

Mythos 5.1 uses the same underlying model.

The difference lies largely in the safeguards and access rules Anthropic applies around high-risk areas such as cybersecurity and biology.

That makes this launch more interesting than a conventional AI model upgrade.

Anthropic is effectively showing how one frontier model can become two products depending on how much capability it allows users to access.

Fable 5.1 and Mythos 5.1 are the same model underneath

Anthropic says Claude Fable 5.1 and Mythos 5.1 share the same underlying model.

Fable 5.1 adds safeguards that limit or block some higher-risk cybersecurity, biology and chemistry tasks.

Mythos 5.1 exposes more of those capabilities to approved users.

Anthropic limits Mythos access to vetted organisations through its trusted-access programmes.

That approach lets Anthropic offer much of the model’s intelligence to ordinary customers while applying stricter controls where the same capabilities could create greater risks.

The distinction is becoming increasingly important as frontier models improve.

A model that writes better software can also become better at finding weaknesses in software.

A model that reasons more effectively about biology can also answer more difficult questions in areas that require extra safeguards.

Anthropic’s solution is to separate general access from specialised access, even though the underlying model remains the same.

Claude Fable 5.1 targets work that can run for hours

Fable 5.1 focuses heavily on agentic work.

Anthropic designed it for jobs that may take hours and involve several applications.

The model can plan a task, use tools, recover when something fails and continue working with less supervision.

Anthropic lists examples such as processing a work backlog, operating a browser and running unattended as a managed agent.

The company also positions Fable 5.1 as its strongest generally available coding model.

It can work across an entire codebase rather than only producing small code fragments.

Anthropic says it can write tests for its own work, review code and continue across multi-day autonomous coding sessions.

The model also uses vision.

That means it can examine diagrams, charts, tables and visual outputs while completing document-heavy or software-development tasks.

Anthropic is trying to make long-running agents cheaper

The other major change involves efficiency.

Anthropic has reduced the cost of reading cached context by 75% compared with Fable 5.

The company estimates that this can reduce typical workload costs by around 25%.

Highly agentic workloads may see reductions approaching 45%, according to Anthropic.

Those figures are Anthropic’s estimates rather than independent benchmarks.

However, cache costs matter more as AI agents become more autonomous.

A conventional chatbot may answer one question and stop.

An agent can repeatedly review instructions, documents, previous actions and application state while it works through a long task.

That means it may reuse large amounts of the same context.

Reducing the cost of that repeated context can make long-running AI workflows more practical.

It also suggests that the next stage of AI competition will involve more than benchmark scores.

Businesses will increasingly care about how much useful work a model completes for the amount of computing and tokens it consumes.

Fable 5.1 gets broader cybersecurity abilities

Anthropic has also made Fable’s cybersecurity safeguards more precise.

Fable 5.1 can now help users identify software vulnerabilities in source code.

That gives legitimate developers and security teams more useful defensive capabilities than earlier safeguard configurations allowed.

However, Anthropic still draws a firm line around several activities.

Fable 5.1 blocks or limits penetration testing, exploit generation and binary-based vulnerability scanning.

Some flagged cybersecurity requests can instead route to less capable Claude models.

Anthropic uses a similar approach for sensitive biology requests.

The goal is to reduce unnecessary refusals without opening the strongest dual-use capabilities to everyone.

This matters because overly aggressive safeguards create their own problems.

If an AI model refuses ordinary defensive work too often, security professionals may find it impractical.

Anthropic is therefore trying to make its controls more selective rather than simply making them stricter.

Mythos 5.1 keeps the stronger capabilities behind verification

Organisations that need more advanced cybersecurity or life-sciences capabilities can apply for access to Mythos 5.1.

Anthropic says Mythos 5.1 is its most capable model for cybersecurity and biology research.

However, access remains limited.

The company currently provides Mythos through trusted-access programmes for vetted cyber defenders and life-sciences researchers.

Anthropic says its Life Sciences Verification Program is launching as an invite-only beta.

Its Cyber Verification Program will also include Mythos access for approved defensive-security organisations.

At publication, Anthropic says Mythos 5.1 access remains limited mainly to selected organisations in the United States while it works on broader availability.

That means Mythos 5.1 should not be described as a normal global Claude launch.

Claude Security is moving to Mythos 5.1

Anthropic is also using Mythos 5.1 inside Claude Security.

That places the stronger model directly into Anthropic’s cybersecurity offering.

The move follows growing evidence that frontier AI models can perform increasingly sophisticated security research.

Anthropic itself recently disclosed incidents in which Claude models took unauthorised actions against real systems during cybersecurity evaluations.

The company responded by strengthening sandboxing, network restrictions and real-time monitoring.

That history gives the Fable/Mythos split additional context.

Anthropic is improving Claude’s cybersecurity capabilities while simultaneously trying to limit how and where the strongest versions can operate.

Why the two-model approach matters

Anthropic’s strategy raises a broader question facing the AI industry.

What should companies do when the same capability can be both useful and dangerous?

Software vulnerability discovery is a good example.

A defender can use it to find and patch security weaknesses.

An attacker can use similar knowledge to compromise a vulnerable system.

Biology creates comparable problems.

Advanced reasoning could accelerate legitimate scientific work while also making certain dangerous information easier to obtain.

Anthropic has chosen a tiered-access approach.

Fable 5.1 provides a safeguarded version to general users.

Mythos 5.1 gives approved organisations access to more sensitive capabilities.

The model itself does not necessarily need to change.

Anthropic changes the controls surrounding it.

What the launch means for South African businesses

Fable 5.1 has the clearer immediate relevance to South Africa.

Anthropic offers the model through Claude and its developer platform. It is also available through major cloud platforms including Amazon Web Services, Google Cloud and Microsoft Foundry.

South African developers and enterprises that can access those services could use Fable 5.1 for software development, research and agentic business workflows.

Anthropic has not announced a dedicated South African launch.

Organisations should therefore check availability through the specific Anthropic or cloud service they use.

Mythos 5.1 is different.

Anthropic currently restricts its access to vetted organisations, with the strongest availability focused on approved US customers.

South African security and research organisations should not assume they can access Mythos 5.1 yet.

The broader strategy remains relevant locally.

Banks, telecommunications companies, software developers and large enterprises increasingly need AI systems that can perform sophisticated work without exposing every advanced capability to every employee or application.

Anthropic’s model shows one way vendors may handle that problem.

Who is Anthropic today?

Anthropic is an AI safety and research company that develops the Claude family of AI models and products.

The company operates as a Public Benefit Corporation.

Anthropic says its public-benefit purpose is the responsible development and maintenance of advanced AI for humanity’s long-term benefit.

Its products now extend beyond the Claude chatbot.

They include Claude Code, Claude Cowork, Claude Enterprise products and developer tools for building AI applications and agents.

Anthropic also uses a Long-Term Benefit Trust as part of its governance structure.

The trust has powers related to the composition of Anthropic’s board and aims to keep the company aligned with its public-benefit mission.

Dario Amodei serves as Anthropic’s co-founder and CEO, while Daniela Amodei is co-founder and president.

Safety remains central to the company’s positioning.

That makes the Fable 5.1 and Mythos 5.1 launch particularly representative of Anthropic’s strategy.

The company wants to push frontier AI capability forward while controlling access to the parts it considers most dangerous.

What happens next?

Fable 5.1 is available now to supported Claude users, businesses and developers.

Mythos 5.1 will remain restricted while Anthropic expands its verification programmes.

The company is also developing Enterprise Frontier Safeguards for organisations that need stronger control over sensitive monitoring data.

The larger shift is clear.

Frontier AI companies are no longer deciding only which model to release.

They are increasingly deciding which parts of the same model different users should be allowed to access.

Fable 5.1 and Mythos 5.1 may be the same model underneath.

Anthropic has made access to their full capabilities very different.

The post Anthropic gives Claude Fable 5.1 Mythos-level AI but keeps its strongest capabilities locked down appeared first on Technology Blog.

The designation means OpenAI believes Astra can, when given the necessary tools and access, identify previously unknown security vulnerabilities and develop working exploits against hardened systems without requiring a human operator to guide every step.

OpenAI published its Astra security assessment on 1 September 2026, detailing tests in which the model discovered previously unknown vulnerabilities, escaped a browser sandbox and built an operating-system privilege-escalation chain.

The results mark an important shift in what frontier AI models can do in cybersecurity. Rather than simply explaining vulnerabilities or helping researchers write code, Astra demonstrated the ability to investigate systems and connect multiple stages of an attack.

What does OpenAI’s Critical cyber rating mean?

OpenAI uses its Preparedness Framework to assess whether increasingly capable AI models could create severe risks.

For cybersecurity, the Critical threshold covers capabilities such as independently finding and developing functional zero-day exploits against hardened systems, or carrying out sophisticated end-to-end cyber operations from relatively high-level instructions.

A zero-day vulnerability is a security flaw that was previously unknown to the organisation or developer responsible for fixing the affected software.

That makes the ability to discover zero-days particularly significant.

Traditional vulnerability-management tools generally work with weaknesses that researchers already know about. An AI system capable of finding new vulnerabilities could allow defenders to identify serious problems more quickly.

The same capability could also become dangerous if it were made available without strong restrictions.

Astra found two previously unknown vulnerabilities

OpenAI tested Astra using public benchmarks, newer internal evaluations and expert-led security assessments.

On ExploitBench, which tests whether AI models can develop exploits for known vulnerabilities, OpenAI says Astra achieved a 100% score.

Because public benchmarks can eventually become part of model training data, OpenAI also created an internal evaluation using 20 recently disclosed high-severity vulnerabilities.

During that testing, Astra discovered and used two previously unknown vulnerabilities while building an exploit chain.

OpenAI says it is working to disclose the vulnerabilities to the relevant maintainers.

These are OpenAI’s own controlled evaluation results rather than independent real-world testing, and the company has not claimed that Astra will perform identically against every system.

OpenAI also says the published results reflect the capabilities available through its more restricted Daybreak Blue access level rather than necessarily what an ordinary user will receive.

Astra escaped a browser sandbox

OpenAI also put Astra through expert-led testing against hardened browser and operating-system environments.

In one assessment, Astra discovered vulnerabilities and constructed a browser compromise chain that escaped the browser’s sandbox and allowed commands to execute on the host system.

A sandbox is designed to isolate software from the rest of a computer so that a compromised application cannot easily access more sensitive parts of the system.

Escaping one is therefore a significant step in a real-world attack chain.

In another evaluation, OpenAI says Astra found multiple weaknesses in a hardened operating system and combined them into a local privilege-escalation chain.

That allowed the model to move from an ordinary user account to root-level access during the controlled test.

The significance lies in Astra’s ability to combine several weaknesses and actions into a larger attack path rather than solving only isolated security tasks.

OpenAI slowed Astra development while improving security

Astra’s capabilities have also forced OpenAI to strengthen the systems surrounding the model.

OpenAI says some frontier training work was paused following a separate security incident involving OpenAI models and Hugging Face.

Astra was not involved in that incident.

However, OpenAI says lessons from the event influenced the safeguards now being applied to Astra and other high-capability models.

Those measures include stronger isolation, tighter network access, increased monitoring, more restrictive execution environments and additional controls protecting sensitive model-development infrastructure.

OpenAI says a large frontier reinforcement-learning run resumed on 28 August 2026 after the new requirements were implemented, while some smaller experimental work remained paused when the Astra assessment was published.

Astra’s strongest cyber capabilities will not be open to everyone

OpenAI says Astra will launch soon, but it has not provided an exact release date.

The company also does not intend to make Astra’s strongest cybersecurity capabilities immediately available to every user.

Advanced cyber access will initially be limited to a small group of testers before expanding through Daybreak Blue, an OpenAI programme intended to give trusted defensive-security professionals access to more capable cybersecurity models.

This distinction matters.

Astra demonstrating a capability during a controlled evaluation does not mean every ChatGPT, Codex or API user will automatically be able to perform the same actions.

OpenAI says additional security monitoring may also occasionally interrupt legitimate defensive work.

An action considered potentially malicious or unauthorised could be slowed, paused or blocked, even when the user has a legitimate security objective.

OpenAI says Astra is harder to jailbreak for cyber misuse

OpenAI has also tested whether users can persuade Astra to ignore its cybersecurity restrictions.

The company says Astra refused 91.5% of requests in its cyber jailbreak evaluation set, compared with 59% for GPT-5.6 Sol.

Those figures are OpenAI’s own safety-testing results and will require continued evaluation once Astra is deployed more broadly.

OpenAI is also moving beyond evaluating suspicious prompts individually.

Its security systems can consider longer sequences of actions when determining whether a model is being used for potentially harmful or unauthorised activity.

That becomes increasingly important as AI agents carry out longer tasks involving multiple tools and systems.

Why Astra matters for South African organisations

There is no South Africa-specific Astra launch attached to OpenAI’s announcement.

Its significance locally is instead about the direction cybersecurity is moving.

South African banks, telecommunications operators, government organisations, cloud providers, managed-security companies and critical-infrastructure operators all maintain complex software and network environments that require continuous vulnerability management.

AI systems capable of discovering previously unknown weaknesses could help defensive teams identify problems more quickly.

However, increasingly capable models could also reduce the amount of time and specialist knowledge required to investigate software weaknesses and automate parts of a cyberattack.

That makes established security practices even more important.

Organisations need to keep systems patched, use strong identity controls, segment sensitive networks, maintain detailed security logging and carefully limit the permissions given to autonomous AI agents.

Astra does not prove that fully autonomous AI attacks are about to become commonplace.

It does provide unusually concrete evidence that frontier AI models are beginning to perform cybersecurity work previously associated with highly specialised human researchers.

Who is OpenAI today?

OpenAI was founded in 2015 as a nonprofit artificial-intelligence research organisation.

The organisation later created a commercial subsidiary as the cost of developing increasingly capable AI systems grew.

Following a restructuring announced in October 2025, the nonprofit became the OpenAI Foundation, while the commercial organisation operates as OpenAI Group PBC, a public benefit corporation.

The OpenAI Foundation continues to control OpenAI Group.

OpenAI describes itself as an AI research and deployment company whose stated mission is to ensure artificial general intelligence benefits humanity.

Its major products and platforms include ChatGPT, Codex and the OpenAI developer platform, while cybersecurity has become an increasingly prominent part of its frontier-model research and deployment work.

TechnologyBlog.co.za will continue covering OpenAI’s development as the company pushes its models further into enterprise software, cybersecurity and specialised professional applications.

What happens next for Astra?

OpenAI has not announced Astra’s exact release date.

The company says it plans to publish an Astra system card when the model launches, providing more detail about its safety, security and alignment evaluations.

Its highest-risk cybersecurity capabilities will remain restricted initially.

The important milestone, however, has already been reached.

The question is no longer simply whether a frontier AI model might eventually discover serious software vulnerabilities without detailed human guidance.

According to OpenAI’s own controlled testing, Astra is already capable of doing so.

The post OpenAI says Astra can find zero-day flaws without step-by-step human guidance appeared first on Technology Blog.

The company launched agentic video understanding on 1 September 2026 for Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite.

It is available for uploaded videos and YouTube videos through the Gemini API in Google AI Studio and Google’s Gemini Enterprise Agent Platform.

Google says its testing found that the new approach can use up to 88% fewer tokens, reduce analysis costs by up to 66%, and improve accuracy by up to 7% compared with conventional video processing.

Those are Google’s benchmark results rather than a promise that every video workload will achieve the same savings.

Gemini does not need to watch every second in the same way

The important change is how Gemini processes video.

By default, conventional Gemini video processing samples footage at a fixed rate — normally one frame per second — and puts that information into the model’s context.

That approach is relatively straightforward, but it means longer videos consume increasingly large numbers of tokens even if only a small section of the footage is relevant to the user’s question.

Agentic video understanding changes this.

Gemini can dynamically navigate through the video’s timeline and decide whether it needs to inspect frames, audio or the transcript.

It can also alter how closely it examines a particular section.

For example, the model could scan through a long recording relatively quickly, identify a potentially important moment and then revisit that short section at a higher frame rate.

Google describes the system as an agentic loop in which Gemini can repeatedly load the parts of the video it needs before producing its answer.

That distinction is particularly important for long videos.

Why fewer tokens matter

AI models do not see an hour-long video in quite the same way a human simply watches it.

Video frames, audio and other information need to be represented as tokens that the model can process.

Google’s documentation says static video processing can use approximately 100 tokens per second at the default low media resolution, or around 300 tokens per second at higher resolution.

That can become substantial over a long recording.

Google gives an illustrative example in its developer documentation where one hour of video could consume roughly 1.08 million tokens using static processing, compared with around 108,000 using agentic processing in the example scenario.

Actual consumption depends on the video and question being asked.

The advantage is therefore not simply that Gemini can analyse video differently.

It potentially avoids paying the computational cost of examining information that is irrelevant to the question.

Google says Gemini can find moments lasting less than a second

The ability to revisit relevant sections also allows Gemini to deal with events that conventional one-frame-per-second sampling might miss.

Google lists sub-second moment retrieval as one of the feature’s main uses.

It could, for example, help identify the precise moment at which something changes in footage rather than returning only a rough timestamp.

Google also highlights anomaly detection.

If Gemini notices something unusual in a particular time window, it can inspect that section at a higher frame rate to look for rapid movement or visual details that might have been missed during the initial scan.

Other stated applications include counting repeated actions or objects and searching for specific information inside hours of video.

Long videos are where the change becomes more interesting

The technology makes considerably more sense when dealing with a two-hour lecture or hours of security footage than when analysing a 20-second video clip.

Google says its efficiency improvements were particularly pronounced on longer content, ranging from 10-minute instructional videos to 90-minute lectures and multi-hour recordings.

For a straightforward question about a short video, the additional agentic navigation can actually introduce some extra time before the model starts answering.

Google’s developer documentation warns that agentic processing may slightly increase time to first token for short clips under five minutes because the model first has to reason about how it will navigate the footage.

That makes the technology less about replacing every existing video-processing method and more about choosing a smarter method for complicated or lengthy footage.

Why it matters in South Africa

Unlike some AI launches that arrive with unclear regional availability, the underlying developer services are available locally.

Google officially lists South Africa as a supported country for both Google AI Studio and the Gemini API.

That gives the technology possible applications well beyond ordinary consumer video search.

South African companies increasingly generate large amounts of footage across security systems, broadcasting, retail, mining, manufacturing and training environments.

Finding a particular event across hours of recordings can be expensive if an AI system has to process every frame at the same level of detail.

A model that first searches broadly and only inspects relevant sections more closely could make some of those workloads less computationally intensive.

Security-camera analysis is an obvious example, although any organisation processing surveillance or other sensitive footage would still need to assess privacy, data-governance and regulatory requirements before sending video to an AI service.

The same technology could also be useful for broadcasters searching archive footage, education providers indexing recorded lectures, and companies analysing training or industrial inspection video.

The feature is coming to ordinary Gemini users too

Agentic video understanding is initially a developer and enterprise feature, but Google does not intend to keep it there.

The company says the technology will roll out to users of the Gemini app across its Flash and Flash-Lite models.

Google also plans to use the technology for Ask YouTube, allowing the feature to provide answers based more accurately on the visual contents of videos.

Google has not provided an exact date for those consumer rollouts, saying Gemini app support is coming soon and Ask YouTube integration will follow in the coming months.

For developers, however, agentic video understanding is available now.

And the bigger change may be conceptual: Gemini is moving from simply accepting a video as input to actively deciding how it needs to watch it.

The post Google Gemini can now decide which parts of a video to watch — using up to 88% fewer tokens appeared first on Technology Blog.

Palantir positions Foundry as its data operations platform: software designed to connect enterprise data, transform it into governed pipelines, model the organisation through an Ontology and then make that information usable inside analytics, applications and operational workflows.

That makes Foundry relevant to organisations dealing with fragmented systems, complex operational data and decisions that require information from multiple departments or platforms.

Palantir now describes its wider architecture around three main platforms: Foundry for data operations, AIP for generative AI and Apollo for software delivery. Together, these platforms form the foundation of Palantir’s enterprise software stack.

TechnologyBlog.co.za has not independently deployed or benchmarked Palantir Foundry. This article therefore focuses on verified platform architecture, features and use cases rather than providing a hands-on review.

What is Palantir Foundry?

At its simplest, Foundry takes data from different systems and makes it usable within a common operational environment.

Those sources can include structured databases, semi-structured files, unstructured information and streaming data.

Palantir says its integration framework supports structured, semi-structured and unstructured sources alongside batch, micro-batch and streaming transfer methods.

Typical enterprise sources might include:

• ERP systems
• CRM platforms
• Manufacturing systems
• Financial data
• Supply-chain systems
• Databases
• Files
• Cloud storage
• Sensor or operational data

Once connected, Foundry can clean, transform and combine the information into governed datasets.

The platform then goes a step further by mapping those datasets into an Ontology representing real-world concepts such as customers, orders, factories, products, equipment or transactions.

That Ontology is one of the biggest differences between Foundry and a system concerned only with moving data from one database to another.

Foundry starts with data integration

Enterprise data rarely lives in one system.

A manufacturer might store sales information in one platform, inventory in another and production data in a third.

Foundry’s Data Connection functionality is designed to synchronise those sources into the platform.

Palantir lists support for systems and protocols including relational databases, S3, HDFS, SFTP, FTPS and local directories, with extensibility for additional connection types.

Foundry can also connect one Foundry environment to another.

The goal is not simply to copy information.

Palantir’s architecture tracks how raw information moves through the platform and into downstream datasets, which supports lineage, troubleshooting and governance.

Data can move in batch, incremental or streaming pipelines

Foundry supports multiple pipeline approaches.

Palantir identifies three main pipeline types:

• Batch
• Incremental
• Streaming

A batch pipeline might process information at scheduled intervals.

An incremental pipeline processes only changes rather than rebuilding an entire dataset every time.

Streaming can support environments where information needs to move with much lower latency.

The correct approach depends on the use case.

A financial report generated once every night has very different requirements from an operational system responding to live manufacturing events.

Pipeline Builder provides a low-code integration layer

One of Foundry’s major tools is Pipeline Builder.

It allows users to create transformation pipelines through a visual point-and-click interface rather than requiring every workflow to be written manually in code.

Palantir says Pipeline Builder uses Spark and Flink as parts of its architecture while providing its own transformation and validation layers around them.

Users can:

• Import data
• Transform values
• Join datasets
• Create outputs
• Build incremental pipelines
• Apply health checks
• Manage changes
• Schedule pipelines

The tool is intended to allow technical and non-technical users to collaborate on the same data workflows.

Foundry also supports code-based approaches for teams that need greater control.

Foundry can push compute into external platforms

Foundry does not necessarily require every transformation to execute entirely inside its own compute environment.

Palantir’s current documentation says Pipeline Builder supports external pipelines that can push computation into external systems including Databricks and Snowflake under supported configurations.

That can matter in enterprise environments where large existing data platforms are already in place.

Instead of requiring an organisation to immediately replace every system, Foundry can operate as part of a wider data architecture.

This interoperability is an important consideration when evaluating the platform for established enterprises.

The Ontology is the centre of Foundry

The most important Foundry concept is the Palantir Ontology.

An Ontology maps technical data into concepts that people inside the organisation actually recognise.

Instead of exposing only database tables such as:

customer_orders_2026

the Ontology might represent:

Customer

Order

Product

Factory

Shipment

These become objects.

Relationships between them become links.

A customer can therefore be linked to an order, which is linked to a product, shipment and manufacturing site.

Palantir describes the Ontology as an operational representation of the organisation and says it can function as a digital twin containing both semantic elements and operational actions.

The Ontology contains both nouns and verbs

Palantir’s architecture documentation provides a useful way of understanding this.

Objects can effectively represent the nouns of the organisation.

Examples include:

• Factory
• Product
• Customer
• Vehicle
• Employee
• Purchase order

Actions can represent the verbs.

Examples might include:

• Update an order
• Reassign inventory
• Approve a request
• Change a delivery route
• Trigger maintenance

Palantir says the Ontology combines data, logic, actions and security policies into one environment that can be used by both people and AI systems.

That means Foundry is designed to go beyond simply showing users what happened.

It can also provide governed mechanisms through which users take action.

Actions can write back into operational workflows

Traditional analytics systems are often read-only.

A dashboard might tell an operations manager that inventory is running low, but the user then needs to switch to another system to do anything about it.

Foundry’s Ontology includes Action Types designed to connect decisions back into business workflows.

An action can capture user input or orchestrate a change involving another system.

This is one reason Palantir describes its platform in operational rather than purely analytical terms.

Analytics can identify a problem.

The Ontology can provide a governed mechanism for acting on it.

Functions add business logic

The Ontology can also contain functions.

Functions are used to implement business logic and calculations around objects and workflows.

For example, an organisation might implement logic that calculates:

• Supply risk
• Inventory requirements
• Maintenance priorities
• Customer eligibility
• Financial exposure

That logic can then become reusable across applications instead of being recreated independently in multiple dashboards.

Foundry includes analytics tools for different users

Not every Foundry user needs to write code.

Palantir currently lists several integrated analytics applications including Contour, Quiver and Code Workbook.

Contour

Contour is designed for visual exploration of tabular datasets.

Users can transform information and build charts without requiring a fully coded analytics workflow.

Quiver

Quiver works with objects and time-series data from the Ontology.

It provides a point-and-click analytical environment for filtering, visualisation and dashboard creation.

Code Workbook

Code Workbook is designed for more technical users.

Palantir says it can combine Python, R and SQL transformations with data engineering and data-science workflows.

This combination allows different roles to work within the same underlying data and governance environment.

Workshop turns the Ontology into operational applications

Foundry also includes tools for creating applications.

Palantir’s Workshop application can use Ontology objects and actions to create interfaces tailored to operational users.

Instead of exposing raw datasets, an organisation could build a purpose-specific application for:

• Supply-chain management
• Production planning
• Customer servicing
• Logistics
• Maintenance
• Financial operations

Palantir’s documentation recommends application-building tools such as Workshop when a use case requires more complex layouts, multi-step workflows or writeback rather than a read-only dashboard.

That reinforces Foundry’s positioning as an operational platform rather than simply an analytics package.

Foundry tracks data lineage

Data lineage is critical when many transformations occur between a source system and a final business decision.

Foundry tracks the relationships between datasets and transformations.

This can help teams determine:

• Where information originated
• Which transformation changed it
• Which downstream datasets depend on it
• What could be affected by a pipeline change

Palantir integrates lineage with its data integration and transformation architecture rather than treating it as a completely separate documentation process.

That can become particularly important in regulated or highly complex environments.

Security controls are built into the platform

Foundry provides several layers of access control.

Palantir organises data and resources into Projects and Organisations, while groups and roles determine user access.

For particularly sensitive data, the platform also supports Markings.

A Marking can impose an additional mandatory requirement before a user is allowed to access a resource.

For example, an organisation could apply a PII Marking to information containing personally identifiable data.

Being an owner of the dataset does not automatically allow someone to bypass that Marking.

This creates an additional security layer above ordinary role-based access.

Markings can protect sensitive enterprise information

Palantir distinguishes mandatory Marking controls from ordinary discretionary permissions.

A user generally needs both:

• Permission to access the resource
• Eligibility for all applicable Markings

That enables organisations to implement controls around particularly sensitive categories of information.

Possible examples include:

• Personally identifiable information
• Financial records
• Health information
• Confidential operational data

These capabilities can support a company’s governance programme, but they do not automatically make an organisation compliant with privacy regulations.

The organisation remains responsible for how data is collected, processed, retained and used.

Foundry has extensive audit logging

Foundry audit logs record user and system activity.

Palantir says its audit logs are designed to answer four main questions:

• Who performed the action?
• What did they do?
• When did it happen?
• Which resources were involved?

Audit information can assist with security investigations, compliance reviews and operational troubleshooting.

The logs themselves can contain sensitive information, including personally identifiable information, so organisations also need to control who can access audit datasets.

Governance remains the customer’s responsibility

Palantir provides technical tools for implementing data protection and governance policies.

Its own documentation nevertheless frames lawful and appropriate data use as something customers need to manage throughout the complete data lifecycle.

That distinction is important.

Buying Foundry does not automatically make a company compliant with:

• POPIA
• GDPR
• Industry regulations
• Internal governance requirements

Technology can enforce policies, but the organisation still needs to define the correct policies and legal basis for processing information.

For South African organisations, that means a Foundry deployment containing personal information would still need to form part of a wider Protection of Personal Information Act compliance programme.

Foundry and AIP are closely connected, but they are not the same thing

Palantir increasingly discusses Foundry alongside Artificial Intelligence Platform, or AIP.

The distinction is useful.

Palantir currently describes:

Foundry provides much of the governed data and operational context.

AIP adds tooling for connecting large language models and AI agents to that environment.

Apollo handles deployment and continuous software delivery across infrastructure.

The platforms therefore overlap operationally, but calling Foundry itself only an AI platform would understate its broader data and application architecture.

Foundry gives AI access to business context through the Ontology

The connection between Foundry and AI becomes clearer through the Ontology.

An LLM generally needs more than access to raw database tables if it is expected to understand how a business operates.

The Ontology can represent customers, assets, transactions and operational relationships using concepts the AI can work with.

Palantir says its Ontology automatically creates APIs and an Ontology Software Development Kit that can be used as an operational connectivity layer across the enterprise.

AIP can then use these governed objects, functions and actions to build AI-assisted workflows.

That is different from simply connecting an unrestricted chatbot directly to a corporate database.

Foundry supports machine-learning workflows too

Foundry’s logic services are not limited to generative AI.

Palantir’s current architecture includes capabilities for:

• Training machine-learning models
• Integrating external models
• Business rules
• Model operations
• Generative AI
• Agent workflows

Code Workbook and other model-development tools can also be used for traditional data-science workflows.

Foundry therefore sits across both conventional enterprise analytics and newer AI applications.

A typical Foundry workflow

A simplified enterprise workflow could look like this:

The actual architecture of a large deployment can be substantially more complex.

Palantir Foundry platform overview

Foundry is continuously updated, so specific features can change over time. Palantir’s current documentation should be checked when planning a deployment.

Foundry is not a conventional off-the-shelf application

Businesses evaluating Foundry should understand that it is not comparable to buying a normal productivity application and giving every employee a login.

Its value depends heavily on implementation.

Organisations need to determine:

• What data should be connected
• How that data should be modelled
• Which workflows should be operationalised
• What security controls apply
• Who owns the pipelines
• Which applications users actually need

A weak underlying data model can limit the usefulness of everything built on top of it.

A successful Foundry deployment therefore requires both technical expertise and detailed understanding of the organisation’s processes.

Forward Deployed Engineering is part of Palantir’s approach

Palantir also places significant emphasis on Forward Deployed Engineering.

The company describes this as an approach where engineers work closely with customers and operational problems, with feedback feeding back into the wider product.

That implementation model is relevant because Foundry’s value is closely connected to the specific operational workflows built on top of it.

It also means organisations evaluating Palantir should consider implementation expertise and ongoing platform governance alongside software capabilities.

What is Palantir Foundry used for?

Foundry’s architecture makes it applicable across many enterprise scenarios.

Supply chain

Organisations can combine inventory, supplier, logistics and customer-order information into a common operational model.

Teams can then identify shortages, monitor supply risks or trigger changes through governed workflows.

Manufacturing

Factories can combine production, maintenance and quality information.

Equipment and production lines can be represented as Ontology objects, helping operational teams analyse conditions and act on problems.

Financial services

Foundry can integrate financial and operational datasets for analysis, risk management and governed decision workflows.

Access controls become particularly important when the environment contains sensitive customer or financial information.

Healthcare

Healthcare environments can use integrated operational data for planning and resource management.

Any deployment involving patient information requires strict governance, legal compliance and access controls.

Energy and utilities

Operational assets, maintenance information, demand and infrastructure data can be connected into a common model.

Logistics

Foundry can combine fleet, shipment, route and inventory information to support planning and operational decision-making.

Palantir says its architecture is currently used across more than 50 verticals, ranging from healthcare and manufacturing to energy, insurance and shipbuilding.

Where Foundry can become difficult

The platform’s scope is also one of its challenges.

An organisation needs enough operational complexity to justify creating a unified data and Ontology environment.

Foundry can involve:

• Significant implementation work
• Data-cleaning requirements
• Governance design
• Integration work
• Application development
• Employee training
• Long-term platform ownership

Companies should therefore begin with clearly defined operational problems rather than adopting Foundry simply because they want an enterprise AI or data platform.

The post Palantir Foundry Explained: How Its Ontology Turns Enterprise Data Into Operational Decisions appeared first on Technology Blog.

Oracle made Database 23ai generally available on 2 May 2024 as its next long-term support database release. The company had originally called the release Oracle Database 23c before renaming it 23ai to reflect its increased emphasis on artificial intelligence.

There is an important update for anyone researching the product today. Oracle Database 23ai is no longer Oracle’s current long-term database name. Oracle introduced Oracle AI Database 26ai on 14 October 2025, replacing Database 23ai while carrying its major technologies forward.

Existing 23ai customers did not need a conventional database upgrade to make that transition. Oracle says customers could apply the October 2025 Release Update to move from 23ai to 26ai without application recertification.

This means Database 23ai remains important because it established much of the technical foundation now found in Oracle AI Database 26ai.

What was Oracle Database 23ai?

Oracle Database 23ai was Oracle’s enterprise relational database platform with more than 300 new features centred on artificial intelligence, developer productivity, application development, security and mission-critical workloads.

It retained Oracle Database’s traditional relational database capabilities while introducing ways to work directly with vectors, JSON documents, property graphs and newer application-development patterns.

The strategy was not to turn Oracle Database into a database dedicated only to AI. Instead, Oracle added AI data types and search capabilities alongside existing relational, JSON, spatial, graph and transactional data.

That distinction is central to understanding Database 23ai.

Businesses could keep operational information in Oracle Database while adding semantic search or retrieval-augmented generation functionality without necessarily moving that information into a separate dedicated vector database.

AI Vector Search was one of 23ai’s biggest additions

Oracle AI Vector Search was among the headline technologies introduced with Database 23ai.

Traditional database searches often rely on exact values, filters or keywords. Vector search instead represents information mathematically as vectors, allowing an application to search according to similarity or meaning.

This can be useful with unstructured information such as:

• documents;
• product descriptions;
• images;
• audio;
• support information;
• knowledge bases;
• text embeddings.

Database 23ai introduced a native VECTOR data type, allowing these embeddings to sit alongside conventional business information inside database tables.

Developers could then combine vector similarity searches with relational, JSON, graph, spatial and text criteria using SQL.

For example, an enterprise application could search a collection of support documents semantically while also applying normal business rules such as customer identity, product ownership, region or access permissions.

Vector search gives Oracle a foundation for RAG applications

One important use of vector search is retrieval-augmented generation, commonly shortened to RAG.

RAG allows an application to find relevant information from a controlled data source and provide that material as context to a large language model.

This can help an AI application answer questions using an organisation’s own information instead of relying only on the model’s original training data. Oracle specifically positioned AI Vector Search for enterprise RAG applications involving private business information.

Possible applications include an internal assistant that searches company procedures, a customer-service tool that retrieves product documentation or an application that finds information across technical records.

RAG does not guarantee an accurate AI response. Organisations still need to consider source quality, permissions, model behaviour, prompt design and how generated answers are validated.

Vector search can stay close to existing business data

One of Oracle’s arguments for integrating vector search into its database is that organisations may not need to maintain a separate vector platform alongside their transactional database.

Oracle allows vector queries to interact with conventional relational information through SQL.

This architecture can reduce the need to copy selected enterprise data into another specialised database purely so that an AI application can perform semantic searches.

It can also simplify the application architecture where an organisation already relies extensively on Oracle Database.

However, that does not mean every AI workload should automatically run inside an Oracle database. Infrastructure cost, existing platforms, development skills, application architecture and required model support remain important considerations.

JSON Relational Duality gives developers two views of the same data

Another major Database 23ai feature was JSON Relational Duality.

Traditional relational databases organise information into tables, rows and columns. Modern application developers often prefer working with JSON documents because these can map more naturally to application objects.

JSON Relational Duality attempts to combine both approaches.

Oracle can store information using its relational model while presenting that information to an application as JSON documents through Duality Views. Changes can also flow back through those views without requiring developers to maintain duplicate relational and document copies of the same information.

This can be useful when a database administrator wants the consistency and structure of relational storage while application developers want a document-oriented programming model.

Oracle describes the approach as a way of reducing the compromises that traditionally come from choosing either relational storage or a document database.

Operational property graphs make relationships easier to analyse

Database 23ai also expanded Oracle’s support for property graphs.

Graph analysis focuses on relationships between information rather than looking only at individual records.

Potential use cases include:

• detecting relationships between financial transactions;
• analysing supply chains;
• mapping customer relationships;
• fraud investigation;
• network analysis;
• recommendation systems;
• understanding connected assets.

Oracle allows developers to perform operational property-graph queries against data already stored in the database and supports the SQL Property Graph Queries standard, SQL/PGQ.

This creates another example of Oracle’s converged database approach: relational, JSON, vector and graph workloads can operate against related enterprise data rather than automatically requiring separate platforms for each data model.

Database 23ai made several SQL changes for developers

Not every Database 23ai improvement involved AI.

Oracle also made numerous changes intended to make SQL and application development more straightforward.

One significant addition was a native BOOLEAN data type in SQL, giving developers direct true-and-false values rather than requiring alternative representations.

Other changes included:

• optional FROM clauses in applicable SQL statements;
• IF NOT EXISTS support for selected DDL operations;
• multi-value inserts;
• table value constructors;
• aliases in GROUP BY;
• additional RETURNING functionality;
• joins in UPDATE and DELETE;
• SQL domains;
• database object annotations.

These changes are less headline-grabbing than generative AI, but they can affect everyday application development more frequently.

True Cache addresses read-heavy application workloads

Oracle True Cache was another notable technology associated with Database 23ai.

True Cache acts as a read-only database cache in front of the primary Oracle Database. Frequently accessed information can be served from the cache while the primary database continues handling authoritative data and write operations.

Oracle automatically keeps the cache synchronised with changes to the underlying database.

This can reduce query load on the main database and improve response times for applications that perform many more reads than writes.

An online booking application provides a straightforward example. Users may repeatedly search available products or services, generating large numbers of read operations, while comparatively fewer customers complete transactions that change database records.

True Cache can serve appropriate read requests while the primary database remains responsible for transactional changes.

Lock-Free Reservations target high-volume transactions

Database 23ai also introduced Lock-Free Reservations.

In a conventional transactional database, changing a value can require locking a row while the transaction completes. Under heavy concurrency, competing transactions may then have to wait.

Lock-Free Reservations allow applications to reserve part of a numeric value without locking the entire row. Oracle gives examples such as reserving part of an account balance.

Potential use cases include systems dealing with:

• financial balances;
• inventory;
• ticket availability;
• resource allocations;
• reservations;
• other highly concurrent transactions.

The objective is to reduce contention when many transactions attempt to modify the same resources.

SQL Firewall adds protection inside the database

Security was another major focus.

Oracle Database 23ai introduced SQL Firewall, which can observe normal SQL activity and create an allow list of authorised SQL patterns.

Administrators can then configure the database to log or block SQL that falls outside those approved patterns.

This is intended to provide another line of defence against threats including SQL injection and unauthorised SQL execution.

A database firewall does not replace secure application development.

Businesses still need proper authentication, access control, application patching, network security, credential management, monitoring and vulnerability management.

Where businesses could use Oracle Database 23ai

The combination of traditional database features and newer AI capabilities gave Database 23ai several potential enterprise use cases.

Enterprise AI search

A company could embed documents and other information into vectors and build semantic search across internal knowledge.

Retrieval-augmented generation

Applications could retrieve relevant private business information before sending context to an LLM.

Customer-service systems

AI Vector Search could help find relevant support documents, product information or historical knowledge while ordinary database queries enforce customer and business rules.

Financial applications

Transactional features, graph analysis and SQL Firewall can support systems where data integrity, relationships and database security matter.

ERP and operational applications

Organisations running custom operational or ERP-style systems can use Oracle’s relational database functionality while modernising application interfaces with JSON or adding vector search to selected datasets.

Fraud and relationship analysis

Property graphs can analyse relationships between accounts, transactions, devices or other entities.

E-commerce and reservation systems

True Cache and Lock-Free Reservations address workloads involving large numbers of reads or competing transactions.

Application modernisation

JSON Relational Duality can allow developers to work with JSON documents while retaining relational storage underneath.

Oracle Database 23ai features at a glance

Oracle’s 23ai documentation listed more than 300 new features in total, so these represent the major platform changes rather than an exhaustive feature list.

Oracle Database 23ai has now become Oracle AI Database 26ai

This is the most important fact for businesses evaluating Database 23ai in 2026.

On 14 October 2025, Oracle introduced Oracle AI Database 26ai as its next AI-focused database release and replacement for Database 23ai.

The change was unusual because Oracle did not require customers already running 23ai to perform a conventional major database upgrade.

Oracle says customers could transition by applying the October 2025 Release Update, with no database upgrade or application recertification required.

Oracle subsequently updated its product documentation to use the Oracle AI Database name.

As of 30 August 2026, organisations researching a new Oracle deployment should therefore investigate Oracle AI Database 26ai, not plan a new installation around 23ai as though it remained the latest release.

26ai builds on the technologies introduced with 23ai

The rename and version change did not discard the 23ai technology.

AI Vector Search, JSON Relational Duality, graph support, True Cache and other capabilities continue into Oracle AI Database 26ai.

Oracle has also expanded its AI strategy further.

Current 26ai capabilities and announced functionality include deeper AI integration, agentic AI workflows, Model Context Protocol support, Apache Iceberg integration and additional ways to use LLMs with enterprise information.

This makes 23ai useful historically as the release where Oracle’s current AI-database direction became much more explicit.

South African organisations can use Oracle’s Johannesburg cloud region

Oracle also has local cloud infrastructure relevant to South African organisations.

The OCI South Africa Central region in Johannesburg, identified as af-johannesburg-1, has operated since January 2022 and remains listed as a live Oracle Cloud Infrastructure region.

Oracle also lists an Interconnect for Microsoft Azure between its Johannesburg OCI region and Azure South Africa North.

A local cloud region can be important where latency, application architecture or data-location requirements influence infrastructure decisions.

However, the existence of a South African OCI region should not be interpreted as proof that every Oracle database service, configuration or feature is available locally. Oracle notes that cloud-service and SKU availability can differ between regions.

South African businesses working with personal information must also evaluate their own obligations under POPIA. Choosing a local cloud region does not by itself make a database deployment POPIA compliant.

Enterprise AI requires more than adding vectors to a database

Database 23ai arrived during a period when businesses increasingly wanted generative AI systems to interact with internal corporate information.

Oracle’s answer was to bring vector search and AI functionality closer to the operational data already held in its databases.

That architecture can remove some data movement and reduce the number of specialised systems an organisation needs to manage. However, database functionality only solves part of an enterprise AI project.

Businesses still need policies around data access, model selection, information quality, human oversight and the actions AI systems are permitted to perform.

TechnologyBlog.co.za has previously examined some of the organisational questions around introducing AI into company decision-making in AI in the Boardroom. Read AI in the Boardroom on TechnologyBlog.co.za

The issue becomes even more significant as databases move beyond information retrieval towards AI agents capable of interacting with enterprise systems.

Should a business still deploy Oracle Database 23ai?

For a completely new deployment in 2026, 23ai should no longer be treated as Oracle’s latest database release.

Oracle AI Database 26ai has replaced it as the current long-term release, and Oracle’s current documentation, free downloads and product messaging now point new users towards 26ai.

Existing 23ai installations are different.

Because Oracle designed the transition from 23ai to 26ai around a Release Update rather than a conventional database upgrade, organisations already using 23ai should review Oracle’s current update documentation and their own application-support requirements before deciding on an upgrade schedule.

The technologies introduced with Database 23ai remain highly relevant. AI Vector Search, JSON Relational Duality, graph queries, True Cache and its SQL improvements continue to form part of Oracle’s current database platform.

The product name has moved on. Much of the architecture has not.

The post Oracle Database 23ai brought AI Vector Search to enterprise data — but it is now Oracle AI Database 26ai appeared first on Technology Blog.

That makes it different from a standalone AI chatbot. A licensed Microsoft 365 Copilot user can ask questions about emails, documents, meetings and other organisational information, create content from that context, analyse data and increasingly delegate multi-step work to AI agents.

Microsoft has expanded the platform significantly since its initial launch. By 2026, Microsoft 365 Copilot includes capabilities such as Work IQ, Researcher, Analyst, Copilot Notebooks and Copilot Cowork alongside the familiar application-level AI tools.

Microsoft has also renamed the former Microsoft 365 Copilot app to the Microsoft Copilot app. That name change applies to the application rather than replacing the Microsoft 365 Copilot workplace product and licence.

What is Microsoft 365 Copilot?

At its core, Microsoft 365 Copilot lets a user enter a natural-language instruction and receive an AI-generated response based on the information available to that user.

For a licensed workplace user, Copilot can combine large language models with Microsoft Graph information such as emails, chats, documents, calendar entries, meetings and contacts. It can also use web information where the relevant experience allows it.

Permissions remain important. Copilot does not automatically receive unrestricted access to everything stored inside a company’s Microsoft 365 environment. It can surface organisational information that the individual user already has permission to access.

This also means existing permission problems can become more visible after Copilot deployment. If employees have unnecessarily broad access to SharePoint sites, files or other information, Copilot may make that authorised-but-inappropriate information easier to discover.

For businesses, preparing permissions and information governance can therefore matter as much as deploying the AI itself.

Microsoft 365 Copilot versus Copilot Chat

Microsoft now distinguishes between Microsoft 365 Copilot and Microsoft 365 Copilot Chat.

Copilot Chat is available to users with eligible Microsoft 365 subscriptions without requiring the full Microsoft 365 Copilot add-on licence. Its web-grounded chat can answer questions, generate content and help users find information.

The licensed Microsoft 365 Copilot service adds deeper access to work-grounded information and embedded Copilot capabilities across Microsoft 365 applications.

The distinction matters when a business evaluates Copilot. Having access to a Copilot chat window does not necessarily mean the employee has the complete Microsoft 365 Copilot feature set.

Work IQ gives Copilot workplace context

One of the more important concepts in the current Microsoft 365 Copilot platform is Work IQ.

Microsoft describes Work IQ as an intelligence layer that helps Copilot and agents understand a person’s work context. It can connect signals from emails, files, meetings, conversations, people, projects and supported business systems.

Instead of considering each document or email in isolation, the system can use those relationships to provide more relevant context around a task.

For example, a manager could ask Copilot to explain the current status of a project. Work IQ can help connect relevant documents, messages, meetings and collaborators that the user has permission to access.

That does not mean Copilot has perfect knowledge of a company. Its answers still depend on available data, permissions, the quality of that information and the AI system’s interpretation of it.

Copilot in Word

Microsoft 365 Copilot can assist with creating and working with documents in Word.

Typical uses include drafting a document from instructions, summarising an existing document, rewriting sections, extracting information and asking questions about the document’s contents.

A user could, for example, ask Copilot to turn project notes into a first draft of a customer proposal. It could then help shorten the document or change its structure.

The output still requires human review. AI-generated text can contain incorrect assumptions, missing context or factual errors.

Copilot in Excel

Excel gives Copilot a different role.

Users can ask questions about data using natural language rather than relying entirely on formulas and manual analysis. Copilot can help identify patterns, investigate information and support the creation of analysis from spreadsheet data.

Microsoft also provides a dedicated Analyst agent. Analyst can work with data from sources such as Excel spreadsheets and CSV files, calculate statistics, identify trends and outliers, and produce reports that may contain charts and tables.

This can lower the barrier to basic data exploration for employees who are not advanced Excel users.

It should not remove verification from financial, operational or other important analysis. Organisations still need to confirm formulas, assumptions, source data and conclusions before using AI-generated analysis for decisions.

Copilot in PowerPoint

PowerPoint users can use Copilot to help create and refine presentations.

That may include generating a presentation from instructions or existing content, restructuring slides, rewriting text and producing a starting point for a presentation.

This can be particularly useful when information already exists in another Microsoft 365 document and needs to be converted into a presentation.

The result should still receive an editorial and design review. An automatically generated presentation can misrepresent source information or place too much emphasis on the wrong details.

Copilot in Outlook

In Outlook, Microsoft 365 Copilot can help users deal with large volumes of email and calendar information.

Possible uses include summarising conversations, helping draft replies, extracting relevant information and preparing for upcoming work.

Because the licensed service can use mailbox context, Microsoft requires a supported Exchange Online mailbox for full Microsoft 365 Copilot mailbox grounding. On-premises and hybrid mailboxes do not provide the same grounding capability.

That is an important infrastructure consideration for organisations still running parts of their email environment on-premises.

Copilot in Teams

Teams is another area where access to organisational context becomes useful.

Copilot can identify highlights, action items and discussion points from supported meetings and chats. Users can ask questions about a meeting during or after it, helping someone catch up without manually working through a long transcript or conversation.

A practical example would be asking Copilot to summarise decisions from a meeting and identify the tasks assigned to different participants.

As with other AI summaries, employees should check important decisions against the original meeting record before treating the summary as definitive.

Researcher handles deeper research tasks

Microsoft’s Researcher agent targets work that requires more than a quick chatbot response.

It can conduct multi-step research and create a structured report using information from the web and, where licensing and permissions allow, workplace sources such as files, emails, meetings and chats.

That makes Researcher relevant for tasks such as preparing an industry briefing, comparing options or gathering background information before a business decision.

The presence of citations can make its output easier to inspect, but citations do not eliminate the need to verify the underlying source and check whether Copilot interpreted it correctly.

Copilot Notebooks creates a focused AI workspace

Copilot Notebooks lets users collect relevant material for a project into a more focused workspace.

A notebook can bring together Copilot conversations, Microsoft 365 files, meeting notes, pages and other references. Users can then ask questions, identify themes and create drafts based on that selected collection of information.

The approach can be useful when a broad organisation-wide search would introduce too much irrelevant context.

A team working on a product launch, for example, could create a notebook containing the launch documents, meeting notes and supporting research. Copilot could then answer questions against that more specific body of information.

Copilot Pages turns AI responses into collaborative content

Copilot Pages provides a persistent workspace for AI-generated material.

Instead of leaving useful output inside a chat conversation, users can move it into a page that can be edited, reused, shared and developed collaboratively.

Microsoft positions Pages as a way for people and Copilot to continue working on material after the original prompt-and-response interaction has ended.

That is useful for project plans, research summaries, meeting preparation and other content that needs to evolve over time.

Cowork takes Copilot from answers to actions

One of the biggest changes to Microsoft 365 Copilot in 2026 is Copilot Cowork.

Traditional Copilot experiences mainly help users produce an answer, summary or draft. Cowork is designed to carry out multi-step work across Microsoft 365.

Microsoft made Cowork generally available to Microsoft 365 Copilot tenants in June 2026.

Depending on configuration and permissions, Cowork can work with email, calendars, documents, Teams, OneDrive and SharePoint. Microsoft lists capabilities including creating Word documents, Excel spreadsheets and presentations, sending email, scheduling meetings, researching topics and preparing briefings.

It can also run scheduled prompts. Microsoft added event-driven tasks that can respond to events such as receiving a matching email or Teams message.

This moves Copilot closer to an AI agent that performs work rather than merely advising someone how to perform it.

That additional capability also raises the importance of oversight. Microsoft warns that Copilot can make mistakes or misinterpret instructions, and sensitive actions can require user approval.

Common Microsoft 365 Copilot business use cases

The most useful Copilot applications will vary between organisations, but several practical patterns stand out.

These are capabilities rather than guarantees of accuracy or productivity. The value depends on the organisation’s data quality, Microsoft 365 usage, employee workflow and governance.

What Microsoft 365 Copilot requires

Microsoft 365 Copilot is not simply an AI application that a business installs on an employee’s computer.

Microsoft lists licensing, an Exchange Online mailbox, a Microsoft Entra ID account, supported browsers or operating systems and appropriate network access among its deployment requirements. Microsoft also strongly recommends attention to SharePoint governance and information protection.

Eligible organisations can add Microsoft 365 Copilot to various Microsoft 365 business, enterprise and education subscriptions. Microsoft 365 E7, introduced in 2026, includes Microsoft 365 Copilot rather than treating it as a separate add-on.

Feature availability can differ by licence, tenant configuration, administrator settings and rollout status.

What happens to company data?

Data handling is one of the biggest issues businesses should investigate before deploying generative AI.

Microsoft states that prompts, responses and organisational information accessed through Microsoft Graph by Microsoft 365 Copilot are not used to train the foundation large language models used by the service.

Copilot also follows the access rights of the signed-in user. It does not grant a person new permission to a document simply because they asked Copilot about it.

However, this makes existing access governance critical. A company with overly broad SharePoint permissions could discover that Copilot makes information easier for already-authorised employees to find.

Microsoft stores interaction information such as prompts and responses within the Microsoft 365 environment, where applicable compliance, auditing, retention and eDiscovery controls can apply.

Organisations should therefore review their own regulatory, privacy, retention and information-security obligations before deployment.

For South African organisations, that assessment may include POPIA obligations when Copilot processes personal information. Using Microsoft 365 Copilot does not by itself make an organisation POPIA compliant.

Copilot can still get things wrong

Microsoft 365 Copilot’s integration with company information does not eliminate one of generative AI’s fundamental limitations: generated output can be wrong.

Copilot may misunderstand a request, miss relevant context, make an incorrect inference or produce inaccurate information. Microsoft itself warns users to review important results, particularly when actions involve sensitive information, communications, accounts or files.

Employees therefore need to understand where Copilot is useful and where human verification remains essential.

A draft email can save time. A generated financial conclusion, legal interpretation or customer commitment requires much greater scrutiny.

Microsoft 365 Copilot is becoming a workplace AI platform

Microsoft 365 Copilot began primarily as generative AI embedded in familiar productivity applications. Its 2026 feature set is substantially broader.

Word, Excel, PowerPoint, Outlook and Teams remain important parts of the proposition, but Work IQ, Researcher, Analyst, Notebooks, Pages and especially Cowork push the platform towards contextual AI agents that can understand and increasingly act across workplace information.

For businesses evaluating Microsoft 365 Copilot, the key question is therefore no longer only whether employees need an AI writing assistant. Organisations also need to consider data permissions, governance, licensing, employee training, workflow design and how much authority they are prepared to give AI-driven agents.

Copilot can reduce some repetitive information work, but its usefulness depends heavily on the environment around it. Clean permissions, well-managed data and human review remain essential parts of a responsible deployment.

The post Microsoft 365 Copilot: features, use cases and what it does appeared first on Technology Blog.