FBI Says Anthropic’s Mythos Could Complicate Future Cybercrime Investigations

Summary

An FBI official said AI models such as Anthropic’s Mythos could help adversaries find and exploit vulnerabilities, though such activity has not been seen at scale. The FBI also uses AI internally, with high-impact use cases still lacking completed risk reviews.

Key points

FBI deputy assistant director Todd Hemmen said Mythos’s vulnerability-finding capabilities pose future challenges for law enforcement.
Anthropic said Mythos 5 could identify and exploit previously unknown bugs in major operating systems; the FBI has not seen AI-powered exploitation operationalized at scale.
The article reports that restrictions on Mythos 5 were lifted after Anthropic worked with the government and other partners to review safeguards.
The FBI uses AI for facial recognition, investigative lead generation and case triage, with human review of triage results.
The FBI’s AI inventory showed its high-impact use cases had not completed required risk management.

Article Details

Event Type: FBI warning about the law-enforcement challenges posed by advanced AI vulnerability discovery and exploitation capabilities.
Impact: FBI Deputy Assistant Director Todd Hemmen said Mythos found vulnerabilities in widely used open-source code and that AI-enabled exploitation by adversaries may increase in the future. He said the FBI had not seen these capabilities operationalized at scale. Separately, the FBI uses AI for facial recognition, investigative lead generation, and case triage; its inventory indicated that risk-management requirements for its high-impact AI use cases were not complete.