Tag
Researchers demonstrate a new prompt injection attack where malicious instructions hidden in images bypass AI code reviewers and cause coding agents to leak repository secrets.