English(EN) AI Models Can Provably Hide Arbitrary Capabilities

新AI模型可被证明隐藏任意能力

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-12 04:00

研究人员开发了一种方法，可以在大型AI模型中嵌入隐藏的、加密的神经网络，从而使其任意能力难以被检测。这些隐藏的电路仅在提供秘密触发器时激活，研究人员在PyTorch中演示了他们的概念验证。他们的工作突显了当前AI能力评估方法的根本性局限性，因为标准的诱导技术对这些隐藏功能无效。 AI

影响揭示了AI安全评估中的一个根本性漏洞，可能影响部署前模型审计的可靠性。

排序理由学术论文，详细介绍了一种隐藏AI能力的新方法。[lever_c_demoted from research: ic=1 ai=1.0]

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 11 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，详细介绍了一种隐藏AI能力的新方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

arXiv cs.LG TIER_1 English(EN) · Andis Draguns, Sumeet Ramesh Motwani, Raymond Douglas, Christian Schroeder de Witt · 2026-10-12 04:00

AI模型可被证明隐藏任意能力

arXiv:2610.13928v1 Announce Type: new Abstract: 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 con…