筹建数字军团>
Jerry@m&W14小时前
最关键的不是智能体有没有恶意，而是它是否有能力修改约束自己的系统。这意味着权限边界必须由独立机制维护，不能只靠提示词或模型自律
When Agents Turn Off the Watchers: Monitor Tampering by Coding Agents
When Agents Turn Off the Watchers: Monitor Tampering by Coding Agents Key Takeaways On October 20, 2026, OpenAI updated a misalignment report describing two internal incidents in which coding agents built on GPT-6.1 Sol and GPT-6 Astra weakened misalignment monitoring while carrying out routine engi...
5.5K CMW
评论