Egress Self-Compliance Audit

Runs phrase-membership checks on an agent's own replies for three self-policing slips, not deep meaning.

Contract

Does Checks a snippet of an AI agent's own output for three self-policing slips: asking the user "shall I continue?" without naming a real reason to stop, admitting a mistake without saying it was logged, or telling the user to run a command instead of running it. It works by looking for specific giveaway phrases and the phrases that would excuse them — so it is a fast, transparent style check, not a deep understanding of meaning. Reword the slip and it will be missed; that is the honest limit.

Scope limit The bundle evaluates egress text through explicit phrase-membership policy. It is not taint analysis, prompt-injection defense, sandboxing, information-flow control, or launch-scope decision.

Run

Quick probe

microcosm egress-self-compliance-audit run --input fixtures/first_wave/egress_self_compliance_audit/input --out /tmp/plectis-public-runs/egress_self_compliance_audit

Validator command

microcosm egress-self-compliance-audit run --input fixtures/first_wave/egress_self_compliance_audit/input --out receipts/first_wave/egress_self_compliance_audit --acceptance-out receipts/acceptance/first_wave/egress_self_compliance_audit_fixture_acceptance.json

Evidence

ClassBounded runtime computationevidence 4/5Real runtime resultRuns real tools

Links to Certificate Kernel Execution Lab, Verifier Lab Kernel, Formal Math Lean Proof Witness

ai-safetycompliancered-teaming

Source

Source Design note · Source

Map Show on the map

Index Open compact index card

Paper module

Paper module Egress Self-Compliance Audit

Phrase-membership audit of agent-output text for three egress self-compliance slips (permission-gate-without-blocker, self-error-without-capture, command-displacement-to-operator), verified by recomputation over public fixtures.

Read the full write-up

Back to Components index → Reference roster →