Popular repositories Loading

1. agentic-misuse-benchmark agentic-misuse-benchmark Public

   Trajectory-level benchmark for prompt injection, policy erosion, intent drift, and coordinated misuse in agentic LLM systems.

   Python 1 1

2. agentguard agentguard Public

   Security linter for AI agent definitions — catches prompt-injection and over-broad-capability holes before they ship. Deterministic, zero-dependency, CI-ready.

   Python 1

3. when-rlhf-fails-quietly when-rlhf-fails-quietly Public

   Research project evaluating silent alignment failures in LLMs under adversarial and high-stakes prompts.

   Python

4. safety-memos safety-memos Public

   Short practical memos on agent safety failures, safeguards, and evaluation design.

   Python

5. agentic-safety-systems-whitepaper agentic-safety-systems-whitepaper Public

   Whitepaper and system design for closed-loop agent safety: trajectory evals, safeguards, release gates, and incident feedback.

   HTML

6. loopforge loopforge Public

   Engineering toolkit for agent loops — lint, scaffold, run, schedule, and eval autonomous loops against a six-block model.

   Python