Agent Security

Microsoft’s Project Perception Puts Agentic Security Into Public Preview

By Kaleido Field Staff · July 28, 2026

Direct answer

Microsoft said Project Perception will enter public preview on August 3 with red-team, blue-team and green-team agent roles coordinated around security context. Its benchmark and cost figures are company-reported and do not establish independent real-world performance.

Microsoft Project Perception security graphic
Image source: Microsoft. Used for editorial coverage of agent security desk.

What happened and why it matters

The reveal is useful as an architecture claim: continuous perception, reasoning and action need inspectable roles and human controls, not merely more automated alerts.

Primary source

Primary reference: Microsoft: Rethinking security for the age of AI. Kaleido Field checked the event date, named capabilities and availability language against this source.

Source check
Source dateJuly 27, 2026
Checked by Kaleido FieldJuly 28, 2026, 11:20 CST
What this source supportsofficial public-preview announcement for agentic security for what is Microsoft Project Perception
What it does not proveIt does not prove a universal product ranking, full regional availability, or performance on every visual intelligence task.

Three stated roles

Microsoft describes red agents that identify potential compromise paths, blue agents that investigate and prioritize risk, and green agents that take corrective actions.

The company frames the three as a closed loop, but its post does not publish an external evaluation of the full loop.

Preview is not general availability

The public preview date matters because a preview is a testable availability state, not evidence of mature deployment across every security environment.

Teams evaluating it still need to review permissions, logging, escalation paths, model behavior, cost, and rollback controls.

How to read the metrics

Microsoft reports 96% on CyberGym and nearly 50% cost savings for one stated configuration. Those are company-reported results tied to a named setup.

They should not be cited as a general ranking of all cyber agents until an independent method and comparable baselines are available.

Evidence boundary

This page reports a dated event from a named primary source. Company specifications and adoption statements remain attributed claims unless independent evidence is cited above.

Reader briefing

Keep the source trail in view.

One concise email when a model, benchmark, or visual-intelligence claim materially changes.

FAQ

What is the practical answer?

Microsoft said Project Perception will enter public preview on August 3 with red-team, blue-team and green-team agent roles coordinated around security context. Its benchmark and cost figures are company-reported and do not establish independent real-world performance.

What source does this article use?

The primary source is Microsoft: Rethinking security for the age of AI. Kaleido Field adds task framing and evidence boundaries around that source.

Where should the user verify the answer?

Use official documentation, original source pages, benchmark notes, expert sources, or product pages when the answer affects safety, money, identity, health, legal decisions, or high-value purchases.