Anthropic’s decision to withhold its Claude Mythos Preview model from public release has stirred concern among senior financial officials after the company said the system had uncovered “thousands of high-severity vulnerabilities” across major operating systems and web browsers.
Concerns over Mythos spread among policymakers and financial officials worldwide, reaching center stage at the IMF and World Bank spring meetings in Washington, the FT reported, as officials turned their focus to the implications of a powerful AI model with advanced cyber capabilities. Officials warned that models such as Mythos could threaten the world banking system by exposing weaknesses in lenders’ cyber defenses.
The emergence of Mythos had rapidly reshaped the policy conversation in Washington, officials said, pushing cyber risk and AI governance ahead of other expected agenda items. Bank of England Governor Andrew Bailey described the development as “a very serious challenge for all of us,” adding that it showed “how fast the AI world moves.” Bailey also said global regulators would need to rapidly evaluate the potential cybersecurity threat to the financial system from Anthropic’s new Mythos model.
Officials said Mythos could shift the balance between cyber attackers and defenders by identifying and combining software weaknesses at a scale beyond human teams, describing it as a “fundamental change in the playing field.”
European Central Bank President Christine Lagarde said the case highlighted how a promising technology could become dangerous if misused. She described Anthropic as “a responsible company” that may now be realizing that while the tool “could be really good,” “if it falls in the wrong hands, it could be really bad.”
U.S. Eyes Limited Path to Mythos
Anthropic has so far restricted Mythos to a limited group of companies under a controlled testing initiative, but some European regulators and executives told the FT they had not yet been able to assess the model directly. Lagarde said officials wanted clear rules for handling such systems, adding, “I don’t think there is a governance framework that is there to actually mind those things. We need to work on that.”
Even so, the United States is preparing limited government access to Mythos, with Bloomberg reporting that the White House is working on safeguards that could allow major federal agencies to use a modified version of the model despite broader concern over its cyber potential. The plan would keep Mythos inside a controlled framework and restrict its use to defensive security work.
