35d ago

Safety Hacking in Constrained Best-of-$N$ Inference-time Scaling

arXiv:2608.22915v1 Announce Type: cross Abstract: Inference-time pipelines often sample multiple outputs, filter them with a learned safety model, and return the proxy-feasible output with the highest learned reward. We show that this composition creates a two-stage failure: an imperfect safety proxy first contaminates the feasible set with unsafe outputs, and reward maximization can then amplify

1 reporttech

Claim audit

No BS check run yet — press ⚖ to extract this story's claims and verify them against independent sources.

All coverage

Safety Hacking in Constrained Best-of-$N$ Inference-time Scaling

rss:arxiv-cscrtech35d ago kagi ↗

arXiv:2608.22915v1 Announce Type: cross Abstract: Inference-time pipelines often sample multiple outputs, filter them with a learned safety model, and return the proxy-feasible output with the highest learned reward. We show that this composition creates a two-stage failure: an imperfect safety proxy first contaminates the feasible set with unsafe outputs, and reward maximization can then amplify