Anthropic Teams Up With Its Rivals to Keep AI From Hacking Everything

Editor's take

Anthropic's Project Glasswing will leverage its Claude Mythos Preview model to enable over 45 organizations, including competitors like Apple and Google, to collaboratively test AI cybersecurity defenses.

This initiative addresses a critical vulnerability: the potential for advanced AI models to be exploited for malicious purposes, from sophisticated phishing attacks to the generation of harmful code. By involving a broad spectrum of industry players, Anthropic aims to build a more robust and shared understanding of AI security risks before they become widespread threats, moving beyond isolated efforts.

Future developments will hinge on the transparency of findings from these collaborative tests and the extent to which actionable security protocols are developed and adopted across the participating entities. The success of Glasswing will be measured by a demonstrable reduction in successful AI-driven cyberattacks, rather than simply the act of collaboration itself.