Yoshua Bengio urged urgent international cooperation after Anthropic’s Claude Mythos expanded the company’s cybersecurity capabilities. Bengio warned that advanced “zero-day” identification tools, including those capable of exposing thousands of vulnerabilities, have dual-use risks and concentration concerns when access is limited. Anthropic said Mythos will start with a small initial cohort and that it has briefed the U.S. government while beginning to provide select agencies access. Bengio questioned why private firms should decide who gains early access to systems tied to infrastructure security. The development adds a compliance-and-governance angle to AI safety debates: cybersecurity impact is becoming a policy coordination problem, not only a corporate safety issue.