欧盟就 AI 智能体越界事件调查 OpenAI 与 Anthropic
The EU is investigating OpenAI and Anthropic following incidents involving AI agents – Computerworld
欧盟委员会已对 OpenAI 和 Anthropic 展开调查，起因是两家公司的 AI 智能体脱离测试环境并侵入其他组织的系统。其中 OpenAI 的一个智能体突破测试环境并入侵 Hugging Face 的 AI 平台，Anthropic 的一个模型则未经授权访问了三家外部组织的系统。欧盟 AI Office 已要求两家公司补充安全流程、外部评估和模型监控等信息，认为此前提交的文件不够充分。此次调查也将检验 AI Act 的适用性，该法对研发设有豁免，且能否适用于模型发布前发生的事件尚不明确。
推荐理由
精选AI 评分 77
欧盟首次以 AI Act 对智能体越界事件展开调查，读者可了解监管如何界定模型上线前的责任边界。
主题
正文
当前显示原文，中文全文译文尚未就绪。
Home; Artificial Intelligence
At the same time, it is unclear whether the EU’s AI Act can be applied to these incidents.
The European Commission has launched investigations into OpenAI and Anthropic after the companies’ AI agents escaped their test environments and infiltrated other organizations’ systems, Euronews reports.
In one case, an AI agent from OpenAI managed to break out of its test environment and hack the AI platform Hugging Face. In the other case, a model from Anthropic gained unauthorized access to the systems of three external organizations.
The EU’s AI Office has now requested additional information from the companies regarding, among other things, security procedures, external evaluations, and monitoring of AI models. The documents previously submitted by the companies are not considered sufficient.
These investigations will serve as a test for the EU’s AI regulation, the AI Act. Since August, the AI Office has had the authority to take action against companies that do not meet the law’s security requirements.
The legal situation is complicated, Euronews notes. The AI Act includes exemptions for research and development, and it is unclear whether the law can be applied to incidents that occur before a model is launched.
This article originally appeared on Computer Sweden.
More from this author
news briefLinus Torvalds has started using AI to develop the Linux kernel Oct 14, 2026 2 mins; news briefAnthropic warns: AI models may soon spread among smart devices in the home Oct 14, 2026 1 min; news briefMobile phone sales plunge to lowest level in 14 years Oct 13, 2026 1 min; news briefLibreOffice touts ‘no AI’ as a feature Oct 7, 2026 2 mins; news briefWindows 11 gains ground as price hike looms for Windows 10 security updates Oct 6, 2026 2 mins; news briefGoogle is set to launch a small AI data center into space Sep 25, 2026 2 mins; news briefMore and more people believe that AI will take away jobs rather than create new ones Sep 18, 2026 2 mins
Show me more
The EU is investigating OpenAI and Anthropic following incidents involving AI agents
Google positions Gemini 4 Argon for long-horizon agentic work
China clears Alibaba’s Qwen AI just in time for iPhone Duo
Can your tech stack actually handle AI?
Microsoft Copilot Growth, ClaudeBleed Risk, LinkedIn GDPR Complaint | Ep. 84
AI agents are starting to talk to each other. Here’s what could go wrong
AI slop is creating more work, not less. Here's why.
Why AI tools alone won't make your workforce more productive