• Anthropic’s AI model Claude Mythos can detect and exploit software vulnerabilities, causing panic across multiple industries.
  • The model found 271 security flaws in Firefox, a 27-year-old vulnerability in OpenBSD, and a flaw in FFmpeg.
  • Anthropic only gave the model to select companies (Google, Microsoft, Apple, JPMorgan Chase), excluding the EU.
  • The EU criticized the lack of access; the European Parliament discussed risks, but Anthropic declined the invitation.
  • The EU’s AI Act could force access from August, but Europe lags in AI development and regulation.
  • European banks demand access, while experts warn similar unrestricted models could appear within months.

Sources & Links