Inference — Monday, May 4, 2026

Monday, May 4, 2026

New Releases

Microsoft Ships Agent 365 for AI Oversight

ecommercenews.asia · May 4

Microsoft made Agent 365 generally available today, positioning it as a single governance layer for AI agents operating across enterprise systems.

Bottom line

Microsoft is treating agent governance as a platform layer, which means every competitor selling point agents now has to answer whether their agents are Agent 365-compatible.

Lens Agents Brings Policy Governance to Cross-Environment AI

helpnetsecurity.com · May 4

Mirantis launched Lens Agents in early access today, providing a governed runtime that connects Claude, Cursor, Copilot, and custom agents to enterprise systems under unified policy and audit controls.

Bottom line

A vendor-neutral agent governance layer entering early access the same day Microsoft ships Agent 365 signals that the market for AI agent oversight is bifurcating into platform-native and infrastructure-agnostic approaches simultaneously.

CrowdStrike Extends Falcon Exposure to Network Devices

globalsecuritymag.fr · May 4

CrowdStrike announced new Falcon Exposure Management capabilities today that extend AI-assisted risk prioritization to network equipment, directly targeting legacy vulnerability management tools.

Bottom line

CrowdStrike is using AI-assisted prioritization as the wedge to pull network vulnerability management into the Falcon platform, shrinking the addressable market for point-solution VM vendors.

Pentagon Signs AI Deals, Excluding Anthropic on Supply-Chain Grounds

ghacks.net · May 4

The Defense Department finalized AI agreements with OpenAI, Google, Microsoft, Amazon, Oracle, Nvidia, SpaceX, and Reflection AI for classified environments, citing a supply-chain risk concern stemming from a contract dispute to exclude Anthropic.

Bottom line

A supply-chain dispute becoming the stated basis for DoD exclusion is the kind of procurement signal that propagates into regulated commercial sectors well beyond defense.

Anthropic Mythos Preview Clears 32-Step Network Attack Simulation

helpnetsecurity.com · May 4

The UK AI Security Institute's testing found Anthropic's Mythos Preview to be the first frontier model to complete a 32-step end-to-end corporate network attack simulation, succeeding in 3 of 10 runs with 73% performance on expert-level tasks.

Bottom line

The same model clearing an offensive attack simulation being deployed for defensive scanning is the clearest illustration yet that frontier AI security capability is genuinely dual-use, not just theoretically.

Funding

Mistral Deepens Singapore Sovereign AI Footprint with HTX and Singtel May 4

Mistral formalized a Strategic Partnership with Singapore's HTX and collaborations with Singtel, NCS, and ST Engineering, signaling that European AI labs are winning sovereign AI contracts by combining open-weight models with local infrastructure commitments that US hyperscalers cannot easily replicate.