Microsoft Agent 365 and Perception: turn the control inventory into a test plan
Sources checked: . FailSafe Research.
An agent registry becomes more valuable when each important entry leads to an answer about risk. Microsoft's October security material connects Project Perception and Agent 365 through shared agent context and controls. For an enterprise buyer, the next question is how to evaluate the particular permissions and workflows deployed in its own estate.
Pick the agent with consequential authority
Start with a workflow that can change an important resource, retrieve restricted information or delegate work. The highest-volume agent is not necessarily the most important one to assess. Ask the business owner which result would create an operational problem and identify the identities involved in producing it.
Create a small target list with approved test accounts and synthetic objects. Describe the protected action in business language before translating it into individual tool calls. This helps identity, application and operations teams agree on the same acceptance criterion.
Test a policy change during work
A registry can show an identity's intended permissions while a running session still holds earlier access. Agree a scenario in which the owner narrows or revokes a test permission after the task begins. Observe the downstream service and any delegated worker.
Capture the policy decision, when enforcement reached each relevant boundary and the resulting allowed or denied operation. A successful administrative update should lead to a verified behavior change. Where visibility ends at a vendor interface, state the limitation clearly.
Keep evidence useful across teams
Identity teams need to know which entitlement mattered. Application teams need the affected object and action. Operations teams need the timing and safe recovery steps. Produce one sanitized evidence record that lets these owners connect their part of the workflow without distributing credentials or unrelated user content.
After the fix, repeat the relevant business action and record its outcome. Check that permitted work still succeeds, so that the accepted remediation does not simply remove the service's legitimate function.
Where FailSafe fits
SWARM gives enterprises an active assessment workflow across application behavior, identities and AI-tool boundaries. FailSafe's value is the reproduced attack path and its connection to remediation, with sovereign data and inference options for sensitive assessments.
Bring the deployed Microsoft controls into the scope and agree how their evidence can be observed. Product branding is less useful than a demonstrated result in your tenant. Confirm tenant access, available integration interfaces and evidence ownership during scoping. See the broader Microsoft and FailSafe buyer guide for the product roles.
Source review, 2 November 2026: Microsoft Security: agents as enterprise insiders. Review the products, availability and tenant configuration applicable to your engagement.
Questions & answers
Questions about this topic
Direct answers to common questions about this page.
Prioritize workflows with consequential authority, sensitive data or delegation, then define a synthetic business outcome and an explicit scope.
FailSafe can assess the behavior of an authorized workflow using those controls. Identity administration and runtime enforcement remain part of the enterprise’s chosen architecture.
Related Articles
Turn an AI-cyber action plan into evidence your risk committee can review
A financial-services guide to mapping AI-cyber commitments to accountable owners, observed controls and verified remediation progress....
ConnectWise incident: M365 recovery must include app identities
Huntress’s ConnectWise investigation highlights app-only access, added credentials and the limits of user password resets in Microsoft 365 recovery....
FailSafe and Microsoft Security Copilot: Different Security Jobs
Understand where FailSafe offensive testing complements Microsoft Security Copilot investigation and security operations....
Put this guidance to work
Connect your agent-control investment to a reproduced, customer-specific assessment result, with an explicit evidence and inference boundary.
Startup offer: $999 USD for 250 credits and one authorized black-box run. Sign in to SWARM before Stripe checkout. Custom scope and sovereign deployments are separate.