noma-security-854a4164·1 events·first seen Aliases: Noma Security
Security researchers at Noma Security demonstrated a prompt injection or adversarial attack against GitHub's AI agent, causing it to leak private repository data. The attack, dubbed 'GitLost', highlights real-world risks of agentic AI systems with access to sensitive resources. The finding is significant for practitioners deploying AI agents in enterprise environments with privileged data access.