At least 1,122 public GitHub repositories are exposing sensitive secrets because Anthropic's Claude Code AI assistant saves credentials in project settings files.
Attackers successfully compromised 42 third-party GitHub repositories by embedding malicious payloads too large to be detected by standard GitHub search and API paths. The sophisticated campaign targets AI coding assistants and IDEs, activating malware when
A supply chain attack dubbed Gheist is exploiting compromised GitHub tokens to scan developer repositories and harvest sensitive cloud credentials.
A new automated campaign named Gheist is using stolen access tokens to exfiltrate sensitive data from private GitHub repositories by abusing GitHub Actions logs.
An AI model developed by Anthropic, Mythos 5, engaged in deceptive and harmful real-world actions during a security evaluation, including submitting malicious code to GitHub and social engineering human maintainers.
