Microsoft Foundry update adds scheduled AI agents and GPT-6 models

By Splitfeed EditorialLast updated Oct 19, 2026, 19:50 UTC

On October 19, 2026, Microsoft released an update to its Foundry platform that brought several notable additions: scheduled AI agents, improved tool use capabilities, and access to GPT-6 models. The same day, reporting emerged that Microsoft had begun classifying AI agents as insider-level entities within its corporate security model, adding what is described as a containment layer to manage the elevated trust and access those agents carry. Separately, Veeam announced an AI-driven undo capability for Microsoft 365 called Intelligent ResOps, which layers onto the broader ecosystem Microsoft is building around autonomous agents.

The combined announcements matter because they move AI agents from experimental tools into infrastructure that sits closer to the core of enterprise operations. Treating agents as insider-level entities is a significant security posture shift: insider threats are among the most expensive and difficult for organizations to manage, and extending that classification to software agents signals that Microsoft views their access levels as comparable to privileged human employees. GPT-6 arriving on Foundry also raises the capability ceiling for developers building on the platform, meaning the agents those developers deploy will be more capable at the same time the security perimeter around them is being restructured. Whether that containment layer is adequate, or merely a rebranding of existing controls, is genuinely in dispute.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically highlight the productivity gains from scheduled agents and GPT-6 access, and frame the insider-level security classification as responsible engineering rather than cause for concern. The Anti-AI camp would likely argue that giving software agents insider-level trust institutionalizes a new attack surface and that a containment layer announced the same day as the expanded capabilities is too little scrutiny too fast. The Middle Ground camp would probably call for independent audits of the containment architecture before organizations roll out scheduled agents at scale, emphasizing that the security model needs to be tested under real adversarial conditions before it can be trusted.

The detail worth watching is how Microsoft describes the technical specifics of that containment layer, either through documentation, a security whitepaper, or developer conference sessions in the weeks following the October 19 release. That disclosure would let security researchers assess whether the insider-threat framing comes with controls strong enough to match the risk it acknowledges.