Anthropic Shifts on Cyber Threats --- Mythos users now allowed to share cybersecurity data in policy change

 

 

“WASHINGTON -- Anthropic recently began letting users of its powerful artificial-intelligence model Mythos share cybersecurity threats with others who may face similar vulnerabilities, modifying its previous stance amid concerns that limiting access to the information could hurt smaller companies.

 

The new policy highlights the challenges facing AI companies that are restricting access to their best models.

 

Anthropic has allowed roughly 50 large companies and organizations managing critical digital infrastructure to access Mythos, which is capable of finding software vulnerabilities much more efficiently than humans, in a program called Project Glasswing. The goal is to let those companies and the government address cyber threats before expanding access and potentially releasing more capable models.

 

The Mythos users initially signed confidentiality agreements to make sure their cyber risk information wasn't shared, a company spokeswoman said. Last week, Anthropic began telling the companies they could share information about cyber threats and Mythos findings with other entities as long as it was done responsibly, she said.

 

"Confidentiality protections were something partners asked for at the outset and were built into agreements partners signed. As the program has matured, we've adapted them to ensure key information can be shared broadly -- including outside the program -- for maximum defensive impact," the spokeswoman said.

 

In one sign of the backlash facing the company, a top Democratic congressman said in a Monday letter to Anthropic Chief Executive Dario Amodei that the company should let Mythos users notify peers in their industries that may face similar challenges, according to a copy viewed by The Wall Street Journal.

 

"No entity should be contractually restricted from warning others, coordinating mitigations, or informing relevant and trusted stakeholders about urgent cyber risks," Rep. Josh Gottheimer (D., N.J.) wrote in the letter. He co-chairs a House Democratic commission on AI and typically backs policies that have the support of many industry executives.

 

Mythos users have already started sharing some information publicly. Cybersecurity firm Palo Alto Networks and Mozilla, the organization behind the Firefox internet browser, recently said the model let them find many more software vulnerabilities than they normally would.

 

Skeptics warn that sharing information more widely could lead to security breaches and cyberattacks if the details fall into the wrong hands. Anthropic last month opened a probe into unauthorized access to Mythos.” [1]

 

1. U.S. News: Anthropic Shifts on Cyber Threats --- Mythos users now allowed to share cybersecurity data in policy change. Ramkumar, Amrith.  Wall Street Journal, Eastern edition; New York, N.Y.. 19 May 2026: A3.