“Yes, company A paid company B to say that company A is safe. Company B is under no duress at all with their findings being what company paid them to find. "Just trust these bros that I've recently paid" You didn't address my question on reflections on trusting trust. For the AI era, there's https://arxiv.org/abs/2610.13928”
AI Models Can Provably Hide Arbitrary Capabilities
Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends on the assumption that hidden capabilities can be elicited. We show this assumption does not hold under adversarial conditions by constructing backdoor attacks that embed encrypted neural networks within host model weights, decrypting and executing them only upon receiving a secret trigger. Using digital lockers, these hidden circuits remain provably hard to elicit and interpret under cryptographic assumptions, even given full access to model weights, extending prior work from hiding fixed strings to arbitrary computations. We implement our proof of concept constructions in PyTorch and evaluate elicitation through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training, using a small chemical reaction prediction task as a proxy for CBRN-relevant capabilities. Our results demonstrate resistance to these methods and, together with our cryptographic guarantees, reveal a limit of elicitation-based capability audits. We release our models to support the improvement of defenses.
候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“Embedded models and triggering: https://arxiv.org/abs/2610.13928 You can add layers to an existing model without massive training costs, which is of course obvious, but ideally you would adjust the weights of a number of layers. Ideally this would be done without knowing the weight decryption inputs, i.e. just a worm behaviour. I recommend the tool hash…”
- AI models can provably hide arbitrary capabilities ↗sbulaev · 2026-10-13 · 282 分