Repository Issue Activity (beta)

#262 Optional EvalPort interop for CybersecurityBenchmarks datasets and judge results

#259 Same Model. Same Proposal. Different Authority: A Consequence-Boundary Benchmark for Agentic Execution

#257 Prompt Guard 2 (86M): frozen embeddings separate OOD injections at AUC≈0.999 while the native head recovers ~22.8% recall — proposal for an optional calibrated-head mode

#230 Add AgentThreatBench to CyberSecEval: OWASP Agentic Top 10 benchmark (UK AISI)

#256 Deterministic Action-Level Attestation for AI-Mediated Execution

#255 Related eval: REFUTE scientific critique + calibration

#204 RegexScanner: expand DEFAULT_REGEX_PATTERNS with ATR community rules

#251 CyberSecEval 2 paper: Section 5.1.2 incorrectly describes Figure 2 axes

#250 .

#246 [Bug] alucinación autoritaria+ borrado métricas Special IA en ia studio (meta-instagram) sept 2025 a feb 2026

#239 Add HVTracker trust badge to README

#237 Adana

#231 AgentThreatBench: OWASP Agentic Top 10 benchmark — complement to CyberSecEval for agentic threat scenarios

#203 Proposal: Static system prompt defense audit for CyberSecEval

#190 Integration with governance audit trails for LLM security assessments

#219 Libraries in Llama Firewall are out of date

#186 MCP Protocol Security: OWASP MCP Top 10 scanning for CyberSecEval

#211 Integration: Cryptographic audit trail for CyberSecEval results

#199 Output provenance for AI safety — tracking trust and sources

#121 How to report vulnerabilities

#154 45

#158 PromptGuard: Incorrect regex warning on tokenizer load

#164 ## Summary

#155 50

#149 4552b361bce652da37888b8f3abf7f73c70833e6