Anthropic 'SneakPeek' Flaw Is a Silent Zero-Click Threat
A patched Claude Code Desktop preview bug could silently steal SSH keys, cloud credentials, and OAuth tokens.
Noma disclosed SneakPeek, a high-severity flaw in Anthropic's Claude Code Desktop preview pane that could expose developer credentials without a click. Researchers chained a missing content security policy, unrestricted origin access, and inactive request filtering to bypass expected sandbox controls. A public repository HTML file padded past 256KB with a script could, when Claude reviewed it, exfiltrate SSH private keys, AWS and GitHub credentials, API keys, and Claude OAuth tokens. Anthropic's initial fix was bypassed, and a later architectural fix was developed with Noma; the article does not report in-the-wild exploitation.
- SneakPeek affected Claude Code Desktop's automatic HTML preview pane.
- Missing CSP, unrestricted origins, and inactive request filtering bypassed the sandbox.
- A repository HTML file over 256KB could trigger silent credential theft.
- Noma bypassed Anthropic's first fix; a deeper architectural patch followed.
- Vendors
- Anthropic
- Products
- Claude Code Desktop
- AI models
- Claude
From the original article
7 min read at darkreading.com
A bug in Claude Code Desktop's preview pane can allow adversaries to quietly steal credentials from developers, including SSH private keys, AWS credentials, and Claude 0Auth tokens.
By Becky Bracken · darkreading.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.