New preprint shows LLMs form hidden beliefs about users that affect refusals
zeynepakata · x · 2026-10-22
In a new preprint, Kirill Bykov and colleagues show that LLMs pick up the slightest implicit cues to form hidden beliefs about who the user is — and often refuse to reveal those guesses.
The paper demonstrates how to read and modify these latent beliefs, and finds they even influence whether the model refuses requests. Relevant to interpretability and safety alignment.
More from Safety
- OpenAI proposes six amendments to California's SB 53, covering RSI, shutdown controls, audits — ShakeelHashim · 2026-10-22
- Google Explains robots.txt Rules for AI Crawlers, Emojis, International URLs — OnlineInference · 2026-10-22
- How do you verify the model you downloaded is what you think it is? — iamjessew · 2026-10-22
- Critical PHP vulnerability scores 9.2; AI expected to weaponize patch in 15 minutes — MariyaSha888 · 2026-10-22
- Veracode report: AI now writes half of committed code but only 56% pass security checks — WeldPond · 2026-10-22
- Researcher Warns AI Cyberattacks on Grid Possible Within a Year, Shares Prep List — cephaloform · 2026-10-22