VEPA converts agent traces into typed events and evaluates them using executable symbolic rules that return allow or refuse decisions.
Formal guardrails can evolve from feedback without surrendering deterministic enforcement
VEPA uses an LLM to propose symbolic policy updates, but activates them only after deterministic checks and sufficient supporting evidence.
Big Tech
Duy Nguyen · Lesly Miculicich · Bhavana Dalvi Mishra · Shuo Li · Krishnamurthy Dj Dvijotham · Mohit Bansal · +2 more
Google Cloud AI Research · Google DeepMind · University of North Carolina at Chapel Hill
Research Digest··2 min read
Nguyen et al.
Why this paper
From Google Cloud AI Research and 2 others
In one line
VEPA adapts formal agent guardrails throughout deployment by verifying LLM-proposed policy patches and activating only evidence-backed rules.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§