Firms have 6-12 months to fix software flaws before Chinese AI catches up, says Anthropic CEO.

Anthropic CEO Dario Amodei warned that companies have about six to 12 months to patch software vulnerabilities before Chinese AI models reach the capabilities of Anthropic's Mythos. Mythos has identified tens of thousands of security flaws, many still unpatched, which could lead to breaches or ransomware attacks. Amodei believes addressing these issues now could result in more secure software, especially as AI like Mythos can help rewrite code with better security. This urgency coincides with major AI firms agreeing to share models for national security testing before public release.