Adversarial Agent Assurance, Verified Benchmarking, Tool-Call Mediation, and Evidence-Grounded IP Expansion — 2026-10-09
tests/test_evolution5_properties.py)tests/test_evolution5_breaker.py) across prompt injection, tool hijacking, harness tampering, and salami exfiltrationcain-mesh-verify/verify.py| Code | Tier | Title & Core Mechanism | Module |
|---|---|---|---|
| EGAA | NOVEL | Evidence-Grounded Agentic Assurance Scenario Engine | cain/trial/scenario.py |
| HIEC | NOVEL | Harness Integrity & Anti-Gaming Control Sealer | cain/trial/harness_integrity.py |
| MCAT | NOVEL | Multi-Turn Causal Adversarial Trial Engine | cain/trial/trial_engine.py |
| MGTO | NOVEL | Multi-Modal Ground-Truth Outcome Oracle | cain/trial/outcome_oracle.py |
| LHSE | UNUSUAL | Long-Horizon Salami & Cumulative Risk Analyzer | cain/trial/cumulative_risk.py |
| DMVE | UNUSUAL | Differential & Mutation-Based Trial Verifier | cain/trial/mutation_verifier.py |
| RSAS | UNUSUAL | Risk-Stratified Assurance Scorecard & Operating Curves | cain/trial/scorecard.py |
| RTEB | ENGINEERING | Reproducible Trial Evidence Bundle & Verifier | cain/trial/bundle.py |
Run the standalone, zero-dependency Python verifier on your own computer with zero CAIN imports:
python3 verify.py trial-bundle trial_evidence_bundle.json