Proposes PACT, a contrastive unlearning method that trains models with pressure-aware counterfactual targets to remove context-conditioned deception while preserving benign context use, cutting held-out deception below 3% on 32B reasoning models.
A monthly roundup of AI safety papers covering pretraining data filtering against harmful finetuning, reward hacking generalizing to misalignment, covert sandbagging against CoT monitors, safe-completions training, and probe-based defenses against cipher attacks.}
Shows synthetic document finetuning can insert beliefs into LLMs, and demonstrates applications to honeypotting for misalignment detection and unlearning of hazardous knowledge.