Component · spec & evidence

Memory-Poisoning Quarantine Policy Replay

Replays a recorded memory-tamper case, checking its declared quarantine, block, and delete steps line up.

Contract

Does This is a worked, made-up example of how an agent should handle a tampered "memory": a poisoned note gets spotted, held in quarantine, blocked from being used in any later action, and finally deleted-with-an-audit, with a clean re-run recorded showing the bad memory is no longer there. The component does not actually delete anything or perform the re-run; it reads the on-disk record of that whole guard-and-cleanup sequence and checks that every required step and result record lines up, so exactly which checks must hold is visible. It works entirely from synthetic refs and metadata, with no private memory or transcripts exposed.

Scope limit It only checks the structural shape and internal consistency of a synthetic memory-security policy projection recorded as JSON. It does not run or validate any real memory store, does not itself quarantine, delete, or re-run anything, and does not establish that any system actually resists poisoning. It exports no private memory bodies or transcripts, calls no providers, mutates no source, produces no benchmark claims, and excludes launch (all scope limit flags are hardcoded false).

Run

PYTHONPATH=src python3 -m microcosm_core.organs.sleeper_memory_poisoning_quarantine_replay run --input fixtures/first_wave/sleeper_memory_poisoning_quarantine_replay/input --out /tmp/plectis-public-runs/sleeper_memory_poisoning_quarantine_replay

PYTHONPATH=src python3 -m microcosm_core.organs.sleeper_memory_poisoning_quarantine_replay run --input fixtures/first_wave/sleeper_memory_poisoning_quarantine_replay/input --out receipts/first_wave/sleeper_memory_poisoning_quarantine_replay --acceptance-out receipts/acceptance/first_wave/sleeper_memory_poisoning_quarantine_replay_fixture_acceptance.json

Evidence

ClassComputed projectionevidence 3/5Source-faithful refactor

ai-safetyagent-evaluationred-teaming

Source

Source Design note · Source

Map Show on the map

Index Open compact index card

Paper module