Anthropic Teams Up With Its Rivals to Keep AI From Hacking Everything
Why it matters: Project Glasswing gives over 40 tech and financial organizations private access to Mythos Preview to mitigate vulnerabilities.
- Anthropic formally announced Mythos Preview, a new Claude model, and established Project Glasswing, an industry consortium, to tackle the cybersecurity implications of advanced AI.
- Project Glasswing includes over 40 tech, cybersecurity, critical infrastructure, and financial organizations, such as Microsoft, Apple, Google, and Amazon Web Services, who will have private access to Mythos Preview.
- Logan Graham, Anthropic's frontier red team lead, emphasizes that the initiative is not just about Anthropic or Mythos Preview, but about preparing for a future where advanced AI capabilities will fundamentally alter modern security paradigms within 6-24 months.
- Dario Amodei, Anthropic CEO, noted that while Mythos Preview wasn't specifically trained for cyber, its proficiency in code makes it highly capable in cybersecurity, highlighting the need for a plan to respond to increasingly powerful AI models from various developers.
- Mythos Preview is capable of advanced exploit development, penetration testing, endpoint security assessment, hunting for system misconfigurations, and evaluating software binaries without source code access, according to Anthropic's Graham.
Anthropic has unveiled Mythos Preview, a powerful new Claude model, and formed Project Glasswing, an industry consortium including tech giants like Microsoft, Apple, and Google, to proactively address the cybersecurity risks posed by advanced AI. This collaborative effort aims to give foundational tech platforms time to mitigate vulnerabilities before such AI capabilities become broadly available, acknowledging that AI's proficiency in code inherently makes it adept at cyber operations.
