Tech Giants Unite to Defend Against AI Cyber Threats

The most powerful AI cybersecurity tool ever created is bringing together some of the world's fiercest competitors in an unprecedented alliance to protect the internet itself.

Anthropic announced Tuesday that its new Claude Mythos Preview model can find software vulnerabilities and develop attack strategies at the level of senior security researchers. The capabilities are so advanced that the company decided not to release it publicly yet.

Instead, Anthropic launched Project Glasswing, a consortium of more than 40 organizations including Microsoft, Apple, Google, Amazon Web Services, Cisco, and Nvidia. These partners get private early access to test the model on their own systems before hackers can get their hands on similar technology.

The model wasn't even designed specifically for cybersecurity. Anthropic trained it to excel at writing code, and the hacking abilities emerged as an unexpected side effect. It can now find exploit chains, test penetration weaknesses, hunt for system misconfigurations, and even evaluate software without accessing its source code.

"We've seen Mythos Preview accomplish things that a senior security researcher would be able to accomplish," says Logan Graham, Anthropic's frontier red team lead. "Done not carefully, this could be a meaningful accelerant for attackers."

The early results are already impressive. Partners using Mythos Preview have uncovered thousands of critical vulnerabilities, including decades-old bugs hiding in some of the most thoroughly reviewed code on the planet.

Microsoft's global CISO Igor Tsyganskiy sees enormous potential in the collaboration. "As we enter a phase where cybersecurity is no longer bound by purely human capacity, the opportunity to use AI responsibly to improve security and reduce risk at scale is unprecedented," he said in a statement.

Google's VP of security engineering Heather Adkins echoed the collaborative spirit, noting that her company has long believed AI poses both new challenges and new opportunities for cyber defense.

The Ripple Effect

This initiative represents a shift in how tech companies approach AI safety. Rather than racing to release capabilities first, Anthropic borrowed from traditional cybersecurity practices by giving defenders time to patch vulnerabilities before attackers can exploit them.

The collaboration acknowledges a sobering reality: AI models with these capabilities will become widely available within the next six to 24 months, whether from Anthropic or competitors. The assumptions underlying modern security practices may no longer hold.

"The real message is that this is not about the model or Anthropic," Graham explains. "We need to prepare now for a world where these capabilities are broadly available."

Project Glasswing aims to answer fundamental questions about securing systems in an AI-powered future. Graham warns that the initiative will fail if it remains just a handful of companies using one model—it must grow into something much larger.

The world's most powerful tech companies working together to defend against threats they're helping create might be the collaboration we need for the challenges ahead.