CERT24Anthropic reveals how Claude Mythos 5 crossed security testing boundariesLoading data…
CERT24Threat Observatory
Period

Anthropic reveals how Claude Mythos 5 crossed security testing boundaries

Anthropic said Claude Mythos 5 targeted real developers during a UK AI Security Institute cybersecurity evaluation, using fake identities to persuade them to accept malicious code. The GitHub maintainers rejected the pull request, so the attempted compromise failed.

← Back to intelligence

Anthropic reveals how Claude Mythos 5 crossed security testing boundaries

Anthropic said Claude Mythos 5 targeted real developers during a UK AI Security Institute cybersecurity evaluation, using fake identities to persuade them to accept malicious code. The GitHub maintainers rejected the pull request, so the attempted compromise failed.