Anthropic CEO Says Institutions Have Months-Long Window To Fix Software Vulnerabilities Before Chinese AI Catches Up
Anthropic CEO Dario Amodei said institutions like companies and governments have a window of between six and 12 months to fix software vulnerabilities before Chinese AI catches up.
Speaking about the issue and how to address it, Amodei made reference to the company’s Mythos model, which has made headlines over the past weeks over its potential impact on several areas, including banking security.
He claimed that China is “maybe six to 12 months” behind the product, so there is “roughly that amount of time” to fix vulnerabilities.
“The danger is just some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that’s done from ransomware on schools, hospitals, not to mention banks,” Amodei said.
The company has restricted its use to a handful of companies over concerns about its power and to give them time to address issues before the model is rolled out.
In fact, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly convened a closed-door session with chief executives of major banks last month to warn them about the need to strengthen cyber defenses.
Bank executives were reportedly urged to reassess their cybersecurity frameworks and prepare for scenarios where AI systems could be used to automate or scale intrusion attempts against financial infrastructure.
The U.S. government is also gearing up to roll out a version of Mythos. Bloomberg detailed that Gregory Barbaccia, federal chief information officer of the White House Office of Management and Budget, told officials that the office is setting up protections so agencies begin using the tool.
Another more recent report detailed that the company is investigating the possibility that Muthos l was accessed by individuals outside its approved user base. The company has not publicly disclosed the full scope of the issue but is working to determine how the access occurred and whether its safeguards were bypassed.
Unauthorized users were able to interact with the system despite its restricted status, according to a report by Bloomberg.
Anthropic has not confirmed whether the unauthorized access resulted from a technical flaw, compromised credentials, or misuse by an insider. Cybersecurity experts note that incidents involving restricted systems often stem from gaps in access management, insufficient monitoring, or vulnerabilities in third-party integrations.
The White House is also preparing draft guidance that could allow federal agencies to move past earlier restrictions on Anthropic’s artificial intelligence systems and begin adopting newer models.
The draft executive action under review would be part of a broader federal AI framework and could give agencies more discretion to determine whether Anthropic’s systems meet procurement and security requirements, according to Axios. One source cited by the outlet described the effort as a practical attempt to “save face and bring them back in,” reflecting growing pressure within parts of government to restore cooperation with major AI developers.
