Proof of Execution: Runtime Verification for Governed AI Agent Actions

2026-07-08T07:23:30Z7d8c190a3154cff895a604a51189a0cf6f14394f5e158d83bccda66b3580515c
ACVAI securityBioSecBench-RefusalHMACLLM-assisted testingPoERAG risksShadowProbeabductive corroborationagent safetyaiAuthZalgorithmic complexity vulnerabilityaudit logauthorization gatewaybiosecurity benchmarkclassifier calibrationexecution attestationforensic synthetic media detectionfuzzinggenerative AI confidentialityprivilegeproduction model stability assessment (FPR calibration)proof of executionruntime verificationstatic analysis

What happened

Collection of new AI-safety and security papers (arXiv 2026-07-08) that advance runtime attestation, tool-call authorization, attack discovery, and risk benchmarks for agentic systems and adjacent domains. Key contributions: (1) “Proof of Execution (PoE)” formalizes attested runtime executions (contract, causal event stream, replay context), defines validator-checkable invariants and semantic guarantees, and proves soundness under cryptographic assumptions; a TypeScript prototype shows low runtime overhead. (2) aiAuthZ introduces an off-host, HMAC-bound authorization gateway with role/argument

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_cr
Record identifier
7d8c190a3154cff895a604a51189a0cf6f14394f5e158d83bccda66b3580515c
Enrichment time
2026-07-08T07:23:30Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.