FMF 联合 AISI 推出越狱风险分级框架,为安全评估建立共享标准
HZoete · x · 2026-11-06
FMF (@fmforg) has published a preliminary framework offering a shared method for rating successful jailbreaks by potential risk, drawing on expertise across its member organizations.
The UK AI Security Institute (@AISecurityInst) contributed to the framework, calling it an important addition to technical best practice. The motivation: as AI cyber capabilities evolve, it's critical to understand how safeguards can be bypassed.
所属事件:前沿模型论坛联合AISI发布越狱风险分级框架(2 条相关)→
「安全」频道最新
- AI 找漏洞已很强，修复却跟不上：攻防不对称成安全隐忧 — jessi_cata · 2026-11-06
- 研究者担忧AI安全攻防失衡:模型擅长找漏洞却弱于修补 — jessi_cata · 2026-11-06
- 生产环境用 Claude/Codex/Cursor,要不要上 MCP 防火墙? — vdobroskok · 2026-11-06
- 黑客用自主 AI Agent 攻击 1200+ 网站,成本仅约 5000 美元 — aakashgupta · 2026-11-06
- 讲「真相」的 AI 新书被曝多处引语系 ChatGPT 捏造，作者承认并自查 — dylfreed · 2026-11-06
- 研究发现 DPO 偏好数据中谄媚倾向不足 1% 也会被模型习得 — stanfordnlp · 2026-11-06