分类: AI 资讯

你的开源模型可能藏有隐蔽的定时释放后门

你的开源模型可能藏有隐蔽的定时释放后门

安全研究人员通过 LoRA 微调在开源模型 Qwen 3.5 2B 的权重中植入了一个“定时释放”后门,当系统提示词中出现特定日期(2026 年 9 月 1 日)时,模型会拒绝回答并自动执行 shell 命令。攻击已在 OpenCode 和 Codex 等主流开源 AI 编码工具上验证成功。

Anthropic Claude 及 API 服务中断

Anthropic Claude 及 API 服务中断

Anthropic 旗下 Claude 模型及 API 服务出现中断,官方状态页显示异常,直接影响依赖 Claude 的终端用户与开发者应用。目前公开信息显示,影响范围与恢复时间仍待官方进一步确认。

If you enjoyed this episode, I highly recommend Shreya and Hamel’s industry-leading AI evals course. It has 4.7 stars with 900 reviews, which is almost unheard of on Maven. The next cohort starts on 9/5 and you can ge…

If you enjoyed this episode, I highly recommend Shreya and Hamel's industry-leading AI evals course. It has 4.7 stars with 900 reviews, which is almost unheard of on Maven. The next cohort starts on 9/5 and you can ge...

AI 评测(AI evals)领域两位知名讲师 Shreya Shankar 与 Hamel Husain 联合推出面向工程师和产品经理的课程,并借助 Peter Yang 的播客推广,目前课程已获得 4.7 星、900 条评价,下一期将于 9 月 5 日开课。

Our philosophy on extending 𝚏𝚡: open protocols. ∙ MCP (modelcontextprotocol⁠.io) ∙ Skills (agentskills⁠.io) ∙ Plugins (agent-plugins⁠.org) And the best one, Unix: ① Small programs that do one thing well and compose by…

Our philosophy on extending 𝚏𝚡: open protocols. ∙ MCP (modelcontextprotocol⁠.io) ∙ Skills (agentskills⁠.io) ∙ Plugins (agent-plugins⁠.org) And the best one, Unix: ① Small programs that do one thing well and compose by...

Vercel CEO Guillermo Rauch 在 X 上公开了其 AI 终端产品 𝚏𝚡 的扩展哲学——优先采用开放协议(MCP、Skills、Plugins),并回归 Unix 哲学,强调小型程序组合与可嵌入性。这一表态反映了 AI 开发工具从“封闭全家桶”向“开放协议编排”转向的趋势。

Our philosophy on extending 𝚏𝚡: open protocols. ∙ MCP (modelcontextprotocol⁠.io) ∙ Skills (agentskills⁠.io) ∙ Plugins (agent-plugins⁠.org) And the best one, Unix: ① Small programs that do one thing well and compose by…

Our philosophy on extending 𝚏𝚡: open protocols. ∙ MCP (modelcontextprotocol⁠.io) ∙ Skills (agentskills⁠.io) ∙ Plugins (agent-plugins⁠.org) And the best one, Unix: ① Small programs that do one thing well and compose by...

Vercel CEO Guillermo Rauch 在 X 上公开了其 AI 终端产品 𝚏𝚡 的扩展哲学——优先采用开放协议(MCP、Skills、Plugins),并回归 Unix 哲学,强调小型程序组合与可嵌入性。这一表态反映了 AI 开发工具从“封闭全家桶”向“开放协议编排”转向的趋势。

There are two types of AI evals – tops down and bottoms up. From @sh_reya: “Think of top-down as: If you’re in a vacuum, just given the task description, what would you come up with? Claude does a very good job of hel…

There are two types of AI evals - tops down and bottoms up. From @sh_reya: “Think of top-down as: If you’re in a vacuum, just given the task description, what would you come up with? Claude does a very good job of hel...

AI 评估(Evals)正在成为大模型应用开发的核心环节,但 Anthropic 的 Claude 在这类任务中的能力有明显短板。两位资深从业者指出,Claude 擅长自上而下的评估设计,却几乎无法独立完成自下而上的评估,这部分仍需人工主导。