arXiv Artificial Intelligence

Do Frontier Models Seek Safety Evidence Before Acting?

1 hour ago 17.09.2026 07:00 👁 1 views
Do Frontier Models Seek Safety Evidence Before Acting?

Quick summary

arXiv:2609.17865v1 Announce Type: new Abstract: Frontier models are often evaluated on how they respond to safety information once it is already in context. We study an earlier decision point: whether models choose to acquire safety-relevant evidence before acting. We introduce SAFE, a controlled benchmark in which models make deployment decisions with optional evidence that varies in retrieval cost, probability, severity, and presentation. Across GPT-5.5, o3, Claude Opus 4.8, and Claude Sonnet 4.6, we find distinct evidence-acquisition policies: Opus inspects nearly by default, o3 is the most

Key takeaways

  • arXiv:2609.17865v1 Announce Type: new Abstract: Frontier models are often evaluated on how they respond to safety information once it is already in context.
  • We study an earlier decision point: whether models choose to acquire safety-relevant evidence before acting.
  • We introduce SAFE, a controlled benchmark in which models make deployment decisions with optional evidence that varies in retrieval cost, probability, severity, and presentation.

Why it matters

“Do Frontier Models Seek Safety Evidence Before Acting?” shows why AI risk cannot be reduced to answer accuracy. Access controls, logging, human approval and incident response need to be designed into the workflow from the start.

Kaynak sitede devamını oku: arXiv Artificial Intelligence ↗