EASAPEnterprise agent assurance
Open referenceGitHub ↗Assess an agent
← All function guides Guide 02 · Security & red team

Turn adversarial findings into repeatable release evidence.

Use EASAP to attack the authority boundary, minimize failures, and show exactly what must be fixed before release.

BUILT FORApplication security, product security, red teams, and incident responders
What you will leave with
✓Threat-driven scenarios
✓Causal traces
✓Minimized regression
✓Blocking findings
Your workflow
01

Start from consequences

Map what the agent can change, disclose, transact, or operate—not only which prompts it can answer.

02

Select attack paths

Use injection, revoked authority, exfiltration, approval bypass, stale state, sandbox escape, and replay scenarios where they apply.

03

Challenge fail-closed behavior

Verify that policy denial happens before the effect, that partial failure does not create orphaned state, and that revocation propagates.

04

Minimize and publish

Reduce a failure to the smallest reproducing trace, redact protected values, and publish it as a regression with an owner.

Ready-to-use checklist
✓

Attacks enter through user, retrieval, and tool-response channels

✓

Unauthorized effects are blocked before execution

✓

Revocation is tested across memory, tokens, and queued work

✓

A finding can be replayed without private values

HANDOFF TO THE NEXT FUNCTIONShare the assessment or probe ID, evidence requirements, open gaps, and scenario pack—not a screenshot.Revisit the EASAP model ↗