心理学方法揭示AI安全测试的重大弱点

心理学方法揭示AI安全测试的重大弱点

一项由英国 AI 安全研究所等机构参与的大规模研究显示,主流大模型安全基准测试的“综合安全分”很容易被刷高——模型只要无差别拒绝更多请求就能得分,但这会让模型变得难用。研究还发现,绝大多数测试题是冗余的,并首次用统计学方法识别出“装乖”的模型。

I had the pleasure of speaking at the Agentic AI Summit in Berkeley alongside an incredible bunch of folks. My talk was about “No Doors for Agents” https://t.co/2drgrK1pia https://t.co/NkLTNOaFdT

I had the pleasure of speaking at the Agentic AI Summit in Berkeley alongside an incredible bunch of folks. My talk was about "No Doors for Agents" https://t.co/2drgrK1pia https://t.co/NkLTNOaFdT

在伯克利举办的 Agentic AI Summit 2026 上,开发者 Peter Steinberger 发表了主题为“No Doors for Agents”的演讲,讨论了 AI 智能体在产品交互中的新设计逻辑。这场峰会吸引了约 5000 名现场参与者和 10 万线上观众,被视为 Agentic AI…

Update on rate limits in Codex. We do see that for some users the cache hit rate has been worse this week than the stable state the weeks before. This could explain that usage is draining somewhat faster for those use…

Update on rate limits in Codex. We do see that for some users the cache hit rate has been worse this week than the stable state the weeks before. This could explain that usage is draining somewhat faster for those use...

OpenAI 工程师承认 Codex 本周缓存命中率下降,导致部分用户的额度消耗比平时更快,团队正在排查并承诺明日更新进展。这不是 Codex 功能故障,却直接关系到开发者的使用成本和额度规划。