AI
AdaGuard: An Adaptive Guard Model with User-defined Policies
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 08:43:02 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 05:18:26 UTC.
Briefings
AdaGuard is a family of language model guard models designed to safely deploy agents using user-defined policies. The project introduces AdaptiveSafety, a dataset of 10,939 training and 1,000 test examples covering policies with 1 to 100 rules. This dataset combines trajectories from multiple sources with policy and behavioral counterfactuals to support learning how to detect violations. The project proposes SafePO, a reinforcement learning algorithm that balances explanatory reasoning with final verdicts. The resulting AdaGuard models include versions of 0.6B, 4B, and 8B parameters, achieving binary accuracies of 89.30% on AdaptiveSafety and 71.82% on DynaBench.
Key points
- AdaGuard is a family of guard models with 0.6B, 4B, and 8B parameters that assess agent trajectories under user-supplied policies.
- The AdaptiveSafety dataset contains 10,939 training examples and 1,000 test examples covering policies with 1 to 100 rules.
Context and limitations
The project repository is available at https://github.com/Yunhao-Feng/AdaGuard. The evidence covers policies with 1 to 100 rules, and the dataset includes trajectories from multiple sources with policy and behavioral counterfactuals.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 08:43:02 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibckv2kmwtpcstv44uc6ecrbjfmac3a5kpdh2snt436rrs6ol7zkq