Steering Might Stop Working Soon — LessWrong

Steering LLMs with single-vector methods might break down soon, and by soon I mean soon enough that if you're working on steering, you should start p…

x Steering Might Stop Working Soon — LessWrong AI Frontpage 68 Steering Might Stop Working Soon by J Bostock 5th Apr 2026 5 min read 13 68 Steering LLMs with single-vector methods might break down soon, and by soon I mean soon enough that if you're working on steering, you should start planning for it failing now . This is particularly important for things like steering as a mitigation against eval-awareness. Steering Humans I have a strong intuition that we will not be able to steer a superintelligence very effectively, partially for the same reason that you probably can't steer a human very

saved by

• Kaustubh Kislay
• Arya Pasumarthi

related reading

• Alignment pretraining could backfire — LessWronglesswrong.com
• llm assistant personas seem increasingly incoherent (some subjective observations) — LessWronglesswrong.com
• Should We Train Against (CoT) Monitors? — LessWronglesswrong.com
• A Mechanistic Explanation of Prompt Injection (and why you should study roles) — LessWronglesswrong.com
• Activation Steering in 2026: A Practitioner's Field Guide | Subhadip Mitrasubhadipmitra.com
• Alignment remains a hard, unsolved problem — LessWronglesswrong.com
• [2605.03907] Steer Like the LLM: Activation Steering that Mimics Promptingarxiv.org
• AI in 2025: gestalt — LessWronglesswrong.com
• Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover — LessWronglesswrong.com
• Reproducing steering against evaluation awareness in a large open-weight model — LessWronglesswrong.com
• You can’t imitation-learn how to continual-learn — LessWronglesswrong.com
• I am worried about near-term non-LLM AI developments — LessWronglesswrong.com