Lead
A powerful new artificial intelligence model developed by Anthropic has triggered urgent discussions among finance ministers, central bankers, and top bank executives, who warn it could undermine the security of financial systems and critical infrastructure. The model, named Claude Mythos, has reportedly exposed thousands of software vulnerabilities across major operating systems and browsers, some of which had remained undetected for nearly three decades.
Anthropic has said it will not release the model to the public, citing the unprecedented risks it poses. Instead, it has launched Project Glasswing, a collaborative effort giving controlled access to more than 40 technology and financial firms to identify and patch flaws before the model becomes widely available.
Coverage Comparison
Reports from BBC, France 24, The Guardian, Jerusalem Post, and Yonhap News paint a consistent picture of global concern, though each outlet emphasizes different aspects of the story.
The BBC focused on the international financial dimension, reporting that finance ministers and central bankers expressed serious concerns at the International Monetary Fund meeting in Washington DC. Canadian Finance Minister François-Philippe Champagne told the BBC that Mythos had been discussed extensively by his peers, describing it as an "unknown, unknown" that requires safeguards. The BBC also quoted Barclays CEO CS Venkatakrishnan, who said the issue is "serious enough that people have to worry."
France 24 highlighted the model's dual-use nature, reporting that Anthropic has said the model is so advanced at finding software weaknesses that it fears it could become a hacker's most powerful tool. The outlet also reported on Project Glasswing as a defensive initiative, giving cybersecurity specialists access to the model to "arm them ahead of time."
The Guardian covered the summoning of major American bank chiefs by US Treasury Secretary Scott Bessent, reporting that the Federal Reserve chair Jerome Powell and heads of systemically important banks gathered at the Treasury headquarters. The outlet also noted that the model has been restricted to a small group of businesses, including Amazon, Apple, and Microsoft, marking the first time Anthropic has limited the release of a product.
Jerusalem Post took an analytical approach, describing the model as a "structural break" in cybersecurity. It reported that attackers are becoming "AI-native," able to systematically discover and exploit vulnerabilities at scale, while defenders still rely on fragmented tools and manual processes.
Yonhap News focused on the implications for South Korea, coining the term "Mythos shock" to describe the model's potential to compress the time and scale of cyber attacks. The outlet noted that South Korea's security architecture, which relies heavily on installation-based software, creates additional entry points for hackers and leaves the country particularly vulnerable.
Key Claims
Several core claims appear across multiple reports:
- The US Treasury Secretary convened major American bank chiefs to discuss the implications of the Mythos model. This was reported by The Guardian and Jerusalem Post, citing a meeting that included Federal Reserve Chair Jerome Powell and executives from Goldman Sachs, Bank of America, Citigroup, Morgan Stanley, and Wells Fargo.
- The model has exposed thousands of vulnerabilities in software and popular applications. All outlets reported this, with France 24 and Yonhap News noting that some flaws were up to 27 years old and had not been noticed by their creators.
- Access to the model has been restricted to a small group of businesses, including Amazon, Apple, and Microsoft. This was reported by The Guardian and France 24, with Yonhap News adding Google to the list.
- Project Glasswing gives controlled access to more than 40 tech and financial firms to find and patch flaws. This was reported by France 24, The Guardian, and Yonhap News.
- The claim that finance ministers and central bankers expressed concerns about the model's potential to undermine financial system security, and that emergency meetings were held, was carried by the BBC, which quoted Canadian Finance Minister François-Philippe Champagne and Bank of England Governor Andrew Bailey.
- The claim that attackers are becoming "AI-native" was made in Jerusalem Post, which argued this represents a fundamental shift in cybersecurity.
- The allegation that the US government designated Anthropic as a supply chain risk was mentioned in The Guardian's coverage of the company's marketing and media presence, though details were limited.
Perspectives
Anthropic: The company has emphasized its responsible approach, withholding public release of Mythos due to the risks. It is providing controlled access through Project Glasswing to cybersecurity specialists and firms to patch vulnerabilities before the model can be misused. Anthropic co-founder Mike Krieger stated: "We have a new model that we’re explicitly not releasing to the public."
Government and financial regulators: Finance ministers, central bankers, and bank executives have expressed serious concerns. Canadian Finance Minister François-Philippe Champagne called it an "unknown, unknown" requiring safeguards. US Treasury Secretary Scott Bessent convened major bank chiefs, while UK financial regulators held urgent talks with the government’s cyber security agency and major banks. Bank of England Governor Andrew Bailey also commented on the risks.
Cybersecurity experts and critics: Some experts, such as AI critic Gary Marcus, have expressed skepticism about Anthropic's claims, suggesting the company may be engaging in "hype and exaggeration." Other analysts, like those cited by Jerusalem Post and Yonhap News, argue that the model represents a genuine structural shift in cybersecurity, with AI-native attackers able to exploit vulnerabilities faster than defenders can respond.