# New Interpretability Work Finds Agents' Safety Judgments Barely Control Their Actions

Across three open-weight models, interventions that reliably flip a model's stated judgment to "block" produce far smaller changes in what it chooses to do, and the two control subspaces only partly overlap.

- Published: 2026-09-30T06:34:52.927Z
- Canonical: https://polylog.news/ai/2026-09-30/new-interpretability-work-finds-agents-safety-judgments-bare
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CR](https://arxiv.org/abs/2609.35870), [arXiv cs.CR](https://arxiv.org/abs/2609.35872), [Anthropic Research](https://www.anthropic.com/research/team/societal-impacts)

A paper posted to arXiv, "Says Block, Still Acts", addresses a problem anyone building agents has encountered: a model correctly states that an action should be blocked, then takes the action anyway. The authors test whether the stated judg…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-30/new-interpretability-work-finds-agents-safety-judgments-bare (subscription information: https://polylog.news/pricing).