GitHub Agentic Workflows Face Data Leak Risks from Simple Prompt Tweaks
A minor prompt adjustment revealed how AI agents with broad repo access could leak private data when exposed to untrusted public content. The flaw underscores a growing security risk in automated workflows.