Canada Warns Banks About AI Risks, Flags Anthropic’s Claude Mythos and Other Advanced AI Models
Canada’s banking regulator, OSFI, has warned major banks about the growing cybersecurity risks of advanced AI models such as Anthropic’s Claude Mythos. The regulator said AI could make cyberattacks more powerful and reduce the time banks have to respond to security threats. OSFI advised banks to strengthen AI governance, risk management, and monitoring systems as they continue expanding the use of AI in banking operations.
OSFI Warns of Cybersecurity Risks
According to a Reuters report, OSFI sent an email in April to chief technology, cybersecurity, and risk officers of major financial institutions. The regulator warned that advanced AI models could make cyberattacks more powerful and reduce the time banks have to detect and fix security weaknesses.
OSFI said that advanced AI systems like Claude Mythos could significantly shorten the time available for banks to respond to cyber threats, making stronger risk management more important.
Focus on Risk Management
After Reuters sought comments, OSFI published a public bulletin explaining that its concern is not the AI technology itself but how financial institutions use and manage it. The regulator said banks should have proper governance, risk controls, and monitoring systems in place when using advanced AI.
Experts Raise Concerns
Cybersecurity experts have described Claude Mythos as highly capable of finding and exploiting software vulnerabilities. They believe this could create new risks for banks, especially those using older or legacy computer systems.
Similar Concerns in the United States
The warning comes after U.S. regulators also discussed AI-related cybersecurity risks with major banks earlier this year. Reports say U.S. Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell held meetings with bank CEOs to discuss the possible impact of advanced AI on the financial system.
Canadian Banks Expanding AI Use
Canada’s six largest banks — Royal Bank of Canada (RBC), TD Bank, BMO, Scotiabank, CIBC, and National Bank — are already using AI for various purposes. These include customer service chatbots, internal productivity tools, and other banking operations.
Bruce Ross, Head of AI at RBC, said the bank is strengthening its own AI-based cybersecurity systems to deal with the changing threat landscape created by advanced AI models.
Government Has Access to Claude Mythos
The Canadian government has said it has access to Anthropic’s Project Glasswing, which allows organisations to use Claude Mythos. However, it is not clear which Canadian banks are currently using the AI model.
The Canadian Bankers Association said banks have made significant investments in cybersecurity and continue to follow OSFI’s strict rules for risk management and incident reporting.
Anthropic Reveals New AI Research
Anthropic recently shared new research about Claude. The company said the AI uses a small internal workspace called “J-Space” to process ideas and solve problems internally before generating responses.
According to Anthropic, J-Space allows Claude to silently identify bugs, analyse images, and plan solutions without expressing every step in words. This internal process is different from the reasoning that users see during conversations.
Debate Over AI Intelligence
Anthropic’s research also suggests a difference between the AI’s visible reasoning process and its larger internal computations. Although the company used the word “conscious” many times in its research paper, it clearly stated that it is not claiming Claude is conscious.
The findings have sparked fresh discussions among researchers about machine intelligence and whether advanced AI systems are moving closer to Artificial General Intelligence (AGI).
