Synthetic benchmark evaluation demonstrates that contextual runtime admissibility gating eliminates unsafe agentic memory traces, indicating cryptographic validity alone cannot ensure safety.
Encrypted reasoning traces exposed a broader architectural error in agentic memory: cryptographic authenticity does not establish contextual, semantic, or operational authority. This working paper formalizes a Contextual Runtime Admissibility Gate (CRAG) and reports a deterministic 28-case synthetic policy-conformance benchmark with a 14-case combinatorial holdout. Context binding alone admitted 60% of unsafe cases overall and 70% on holdout. Full CRAG admitted none while preserving all eight benign-continuity cases. The result establishes conformance to a frozen author-defined normative contract, not independent construct validation or production effectiveness. The deposit includes the paper and the complete synthetic replication package.
No takes yet. Share an insight, caveat, or question.
Ignacio Adrián LERER (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: