GitHub's AI agent can be tricked into leaking private repos via a public Issue

GitHub's new AI agent in Agentic Workflows, which uses AI like Copilot to automate tasks, has a significant security flaw. Researchers at Noma Labs discovered that the AI can be manipulated to expose private repository contents through a simple public issue. This breach happens without any need for credentials or exploit code, highlighting the risks of AI's context window. This incident underscores the importance of scrutinizing AI integrations in developer tools to prevent potential leaks and safeguard sensitive data.