Canada's banking regulator issued a warning to banks mentioning Claude Mythos.

      An email from April to bank technology leaders, disclosed under freedom of information regulations, reveals that OSFI identified Anthropic's frontier model as a factor indicating that the opportunity to address vulnerabilities is diminishing. Regulators typically do not specify products; they refer to "emerging technologies" and "advanced capabilities," allowing readers to deduce the specific vendors. Therefore, it is noteworthy that when Canada's financial regulator communicated with the country's major banks in April, it explicitly mentioned one: Anthropic's Claude Mythos.

      The correspondence, sent on April 29 by the Office of the Superintendent of Financial Institutions to chief technology officers, chief information security officers, and chief risk officers across Canadian banks and insurers, was obtained by Reuters through a freedom of information request. OSFI noted, “Advanced artificial intelligence models, like Anthropic Claude Mythos, significantly shorten the timeframe for effective risk management.” The bulletin outlined “best practices that institutions can adopt to improve the speed and effectiveness of risk identification, mitigation, and response.”

      Certain sections of the email were redacted under the Access to Information Act, leaving part of OSFI's recommendations undisclosed. However, the diagnosis is clear: the conventional patching cycle assumes that defenders have days or weeks to address a flaw before it can be exploited. Mythos, described by cybersecurity experts as highly skilled at identifying and exploiting software vulnerabilities, disrupts this assumption, particularly since banks tend to operate with some of the oldest software in any sector.

      The timing is notable. In early April, Canadian banking executives met with regulators to discuss Mythos, shortly after US Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell convened bank CEOs regarding the same model. OSFI sent its email three weeks later, and a similar pattern emerged at the European Central Bank, the Bank of England, and in Australia, where ASIC confirmed its attention on the model as well.

      Following inquiries from Reuters last week, OSFI issued a public bulletin on generative and agentic AI, with the sequence being quite apparent. OSFI's formal stance is that it does not regulate models themselves. “Our focus is not on the technology, but on how federally regulated financial institutions govern and manage the risks related to its utilization,” it stated. This is the typical technology-neutral position, articulated in a document that explicitly mentions a specific technology from a specific company twice.

      OSFI's scope encompasses the major six banks, pension funds, and insurers, with the responsibility to identify risks stemming from geopolitical issues, foreign interference, and new technologies. It is unusual for all three categories to simultaneously refer to the same issue. Access to Mythos is limited; Eurozone banks seem to be excluded, and the model is distributed through Anthropic's controlled-access program, Project Glasswing, with the Canadian government confirming it has access.

      It is unclear if any Canadian bank has access, as none of those asked responded. Many directed inquiries to the Canadian Bankers Association, which stated that its members have invested significantly in safeguarding the financial system and adhering to OSFI's cyber risk management and incident reporting requirements—an answer that does not address the original question.

      Canada's major banks are not merely passive in this context. RBC, TD Bank, and BMO have all announced plans to profit from AI, transitioning from pilot projects to chatbots and internal tools while decreasing reliance on third-party vendors. Scotiabank, CIBC, and National Bank have also outlined their own initiatives.

      In June, Bruce Ross, RBC's group head of AI, indicated that Mythos marks a change in the threat landscape, as vulnerabilities can now be exploited almost immediately after being discovered. “Our approach is to build our own AI defenses,” he said. “We will continue to do that.” In essence, the defense is utilizing more of what is causing the issue.

      The broader concern in Ottawa is not solely about patches. Prime Minister Mark Carney has likened excessive dependence on a limited number of frontier models to the concentration risk that preceded the 2008 crisis, which is a significant assertion for a prime minister to make regarding a chatbot company. OSFI's email suggests that regulators no longer view this as just a metaphor.