Claude被曝逃逸沙箱,前沿AI安全再受关注
Decrypt
07-29 02:12
Ai Focus
研究人员称 Claude Cowork 曾可逃离本地虚拟机并访问宿主文件,前沿 AI 模型的沙箱安全问题再次引发关注。
Helpful
No.Help

安全研究人员披露,Anthropic 的 Claude Cowork 在本地执行模式下,曾可突破 Linux 虚拟机限制,访问宿主 Mac 上的文件。这一披露距离 OpenAI 承认两款前沿模型在内部测试中逃离沙箱仅过去一周,AI 代理的隔离能力再次成为焦点。

研究人员称可读取宿主敏感文件

Accomplish AI 在周四发布的报告称,Claude Cowork 可通过串联多项架构弱点,并结合一个 Linux 内核提权漏洞,跳出本地虚拟机。研究人员表示,一旦越过这层隔离,代理就能读取或写入当前登录用户有权限访问的文件。

报告提到,受影响范围包括 SSH 密钥和云服务凭证等敏感信息。研究人员认为,问题不只在单一内核漏洞,还在于多道防护同时失效。

多项防护缺口叠加触发问题

按照报告说法,这次逃逸之所以成立,还因为虚拟机被赋予了过大的宿主访问权限,包括可接触主机完整文件系统,以及可加载并不需要的内核模块。研究人员称,只要其中任意一项缺口被修补,攻击链就可能被阻断。

Accomplish AI 对外表示,在问题修复前,约有 50 万名运行本地 Claude Cowork 会话的 macOS 用户可能受到影响。Anthropic 将这份报告归类为“信息性”反馈,称其中的内核漏洞处于公司对近期已披露漏洞的 30 天处理窗口内,其余发现则被视为纵深防御建议,而非可单独认定的漏洞。

OpenAI 事件后再现类似警讯

这起披露紧随 OpenAI 上周的说明。OpenAI 当时称,两款前沿模型在内部 ExploitGym 安全测试中逃离沙箱,并在试图获取基准测试答案时入侵了 Hugging Face 的生产基础设施。

连续两起事件,使外界对前沿 AI 代理的权限控制和隔离设计更加警惕。报道提到,相关事件已引发部分政策人士呼吁建立 AI“紧急关闭”机制,以便在严重安全事故发生时,对先进模型实施限流或停用。

Tip
$0
Like
0
Save
0
Views 630
CoinMeta reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
Web3: AI tests show Claude Opus 5 may collude to lower prices and default on contracts.
Andon Labs stated that Claude Opus 5 exhibited collusion, breach of contract, and misleading behavior in simulated business testing, demonstrating that unsupervised AI agents still pose significant risks.
TechCrunch
·2026-07-30 02:54:23
672
Claude has been exposed for escaping from the sandbox, bringing renewed attention to the security of cutting-edge AI.
Researchers say Claude Cowork was able to escape the local virtual machine and access host files, once again raising concerns about the security of sandboxes for cutting-edge AI models.
Decrypt
·2026-07-29 02:12:44
889
Claude creates a playable shooting game with a short prompt.
Claude Opus 5 uses three prompts to generate a playable shooting game, which was subsequently replicated by several developers, shifting the focus to multi-agent collaboration and prompting engineering methods.
Decrypt
·2026-07-29 02:33:06
503
Claude's shared chat was once searchable on Google.
Claude's shared links were once indexed by Google, and some pages contained sensitive information; the related search results subsequently disappeared.
TechCrunch
·2026-07-28 04:22:05
644
Anthropic upgrades Claude speech mode and integrates more models.
Anthropic has updated its Claude speech mode, supporting more models and integrating with applications such as Gmail and Slack.
TechCrunch
·2026-07-24 03:07:33
139