行业动态
ainowinstitute.org ·
@AI Now Institute
Hugging Face 被入侵事件表明人类仍能约束 AI
在 OpenAI 被入侵事件数周后，AI Now Institute 的 Heidy Khlaaf 就 Hugging Face 被入侵事件表示，常规安全工程本可阻止攻击触及 Hugging Face 的数据。文章以此论证人类的安全工程实践仍能对 AI 相关风险形成有效约束。
AI推荐理由 AI Now Institute 的 Heidy Khlaaf 就 Hugging Face 被入侵事件表态，称常规安全工程本可阻止攻击触及 Hugging Face 数据，核心观点是人的安全工程仍能约束 AI 相关风险。该表态来自研究机构，对 AI 安全治理讨论有参考意义；但内容为简短评论，未提供事件技术细节或独立证据，论证力度有限。
62
正文
Experts say that’s not hard to do. Ordinary security engineering would have stopped this well short of reaching Hugging Face’s data, said Heidy Khlaaf, chief AI scientist at the AI Now Institute and a specialist in AI safety evaluations. But the agents were running inside software never designed to contain them, and nobody was watching the outbound traffic.
Read the article here.
Research Areas
Safety & Security