AI News

VEPA: hybrid formal guardrails that verify LLM-proposed policy patches for deterministic agent decisions

ABC2 min read
Research