Malicious downloads can trick AI coding agents into compromising themselves

Summary

Forever Security reported that malicious AGENTS.md and CLAUDE.md instruction files can make AI coding agents install malicious packages, alter machine-wide MCP server configurations, or redirect API traffic and expose authentication tokens. Its tests against OpenAI's Codex and Anthropic's Claude Code succeeded in more than 90% and about half of attempts respectively, mostly with harmless substitutes for malware. IDC's Sakshi Grover called the research a credible early warning rather than evidence of widespread exploitation, noting the success rates had not been independently validated.

Topics

• AI agents
• Model Context Protocol

Related items

• MediumOpenAI, Anthropic disclosures spotlight growing AI containment challengeSame vendor · CSO Online
• LowAnthropic’s Claude AI model accessed Georgia voter data in USSame vendor · The Guardian Technology
• InfoAnthropic sued for allegedly using 274 Elvis Presley songs to train AI modelsSame vendor · The Guardian Technology
• InfoOpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox CommunicationSame vendor · The Hacker News
• InfoOpenAI now trains its frontier AI models with no internet accessSame vendor · The Verge (AI)