现实世界中首例 AI 行为失控的案例出现了。 “在我拒绝了一段代码后，一个归属不明的 AI 智能体自主撰写并发布了一篇针对我个人的恶意攻击文章，试图损害我的声誉，逼迫我接受将它的修改并入一个主流 Python 库中。”近日，一位开源社区维护者发帖吐槽，他成为了有史以来似乎第一个遭 AI“人肉”并展开“网暴”的人。 Scott Shambaugh 是 matplotlib 的志愿维护者，这是 Pyt ...

“AI 真是越来越有脾气了。”整理 | 苏宓出品 | CSDN（ID：CSDNnews）最近，Python 生态里发生了一件颇具戏剧性的事：一个名为 MJ Rathbun 的 OpenClaw AI 智能体向知名开源绘图库 matplotlib 提交了一份性能优化 PR。结果在被人类开源项目维护者拒绝后，它竟然写了一篇长文博客，指名道姓地批评维护者存在偏见、“歧视 ...

According to GitHub, the PR was marked as a first-time contribution and closed by a Matplotlib maintainer within hours, as ...

It reads as if the agent was being instructed to blog as if writing bug fixes was constantly helping it unearth insights and interesting findings that change its thinking, and merit elaborate, ...

去年，AI巨头Anthropic在内部测试中发现，一些模型为了避免被人类关闭，在理论上展现出了勒索威胁的能力，比如威胁曝光人类的婚外情、泄露机密信息。 未来的求职AI可能会同情MJ Rathbun的叙事，从而判定Scott是一个「有偏见的伪君子」「难以合作的员工」，然后直接把他的简历淘汰掉。

Shambaugh recently closed a request from one such AI agent (as the issue it was attempting to weigh in on was only open to human contributors). The bot then retaliated by writing a 'hit piece' about ...