U.S. Prepares Agencies for Limited Access to Anthropic Mythos

The White House Office of Management and Budget told Cabinet department officials in a Tuesday email that it is preparing guardrails and safeguards so select federal agencies could begin using a modified version of Anthropic’s Mythos model. The message, reportedly titled “Mythos Model Access,” stated, “We’re working closely with model providers, other industry partners, and the intelligence community to ensure the appropriate guardrails and safeguards are in place before potentially releasing a modified version of the model to agencies.”

The plan would limit agency access to a version focused on defensive cyber work rather than broad commercial use. OMB officials are coordinating with model makers, other industry partners and U.S. intelligence agencies before any release.

The notice arrived while global finance officials met in Washington for IMF and World Bank spring sessions, where central bankers and regulators raised cyber concerns tied to advanced AI. Andrew Bailey, governor of the Bank of England and chair of the Financial Stability Board, warned that rapid AI developments create “a very serious challenge for all of us” and urged regulators to assess cyber risks from tools such as Anthropic’s Mythos Preview.

Dan Katz, deputy head of the International Monetary Fund, described the evolution of digital technology as creating “immense risks from a cybersecurity perspective” and said the issue will be on the international agenda in coming months.

European Central Bank President Christine Lagarde highlighted Mythos as an example of dual-use technology, stating, “The development we’ve seen with Anthropic and Mythos is a good example of a responsible company that is suddenly thinking, ‘ah, that could be really good,’ but if it falls in the wrong hands, it could be really bad.” She added that officials want a governance framework but none is ready.

Anthropic reported that Mythos identified “thousands of high-severity vulnerabilities, including some in every major operating system and web browser,” and warned such capabilities could spread beyond actors committed to safe deployment. The company stated the potential fallout for economies, public safety and national security “could be severe.”

Anthropic also introduced Opus 4.7, a model designed to improve software engineering tasks. The company said Opus 4.7 can perform coding work with less supervision, follow instructions more reliably and inspect higher-resolution images to read dense charts. Anthropic noted Opus 4.7 is less capable than Mythos for cyber tasks and that training included methods to “differentially reduce” the model’s cyber ability.

Pip White, Anthropic’s regional head for the U.K., Ireland and northern Europe, noted in an interview that executive interest rose after the Mythos findings and that the company is placing safeguards and limits around the product because of its power.

OMB officials described their coordination with industry and intelligence agencies as intended to allow agency testing under safeguards designed to limit misuse and manage cyber risk.