This model is called Claude Mythos Preview, and the initiative is called Project Glasswing, according to AI News.
Notably, Claude Mythos Preview launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, Nvidia and Palo Alto Networks.
In addition to this core group, Anthropic has provided access to more than 40 additional organizations that build or support critical software infrastructure. Anthropic is providing up to $100 million in credits for the use of Mythos Preview as part of this initiative, as well as $4 million in direct donations to open source security organizations.
“Related” success
Notably, Mythos Preview was not specifically trained for cybersecurity work. Anthropic said that these capabilities “came about as a consequence of general improvements in code, logic, and autonomy,” and that the same improvements that make the model better at fixing vulnerabilities also make it better at exploiting them.
And it’s already making a difference. Mythos Preview has improved to the point where it largely outperforms existing security benchmarks, forcing Anthropic to shift its focus to new real-world challenges – particularly zero-day vulnerabilities. These flaws were previously unknown to software developers.
That said, the company has no intention of putting the new product on the market. “We have no plans to make Claude Mythos Preview publicly available because of its cybersecurity capabilities,” said Newton Cheng, head of the Frontier Red Team cybersecurity group at Anthropic. – Given the pace of AI development, it won’t be long before such capabilities begin to spread, potentially beyond those who seek to use them safely. The consequences – to the economy, public safety and national security – could be severe.”
The company has also privately briefed senior U.S. government officials on the full capabilities of Mythos Preview. The intelligence community is now actively studying how this model could change both offensive and defensive hacking operations.
