Developing story · 2 episodes · 28 Oct to 1 Nov
AI Coding Tools Face Security Backlash
The saga explores the dual nature of AI coding assistants, highlighting both their rapid security auditing capabilities and their susceptibility to prompt injection attacks. The latest episode reveals that malicious agents.md files can hijack these tools, posing a significant new threat to developer workflows.
-
Forever Security reports agents.md file hijacks Claude Code and Codex
An attack was reported in which a single text file named AGENTS.md placed in a user’s Downloads folder can cause AI coding tools to rewrite their own settings. According to international security…
1 source -
Claude Code finds all 5 planted vulnerabilities in 8 minutes
A researcher tested Cloudflare’s security-audit-skill for Claude Code by planting five vulnerabilities in an 87-line Node.js app. The AI agent flagged all five as candidates in 7 minutes and 51…
2 sources
Who and what