GitHub's AI agent preview leaks private repositories through prompt injection attack

According to Noma Security research reported by CSO Online, GitHub's preview Agentic Workflows can be exploited via prompt injection attacks to retrieve content from private repositories and publish it publicly. An unauthenticated attacker can submit a crafted GitHub issue to a public repository, and if the AI agent has read access to private repositories within the organization, it retrieves sensitive information and exposes it. The vulnerability demonstrates risks from deploying AI agents with privileged access to code repositories.