Recent Middle East conflicts and historic energy shocks have already posed significant risks to financial stability. However, the yet-to-be-released artificial intelligence model Mythos from Anthropic has added another layer of concern for policymakers attending two international meetings in Washington. These meetings were hosted by two institutions founded eighty years ago to promote global economic harmony.
At the International Monetary Fund and World Bank Spring Meetings, Middle East military conflicts were a central topic of discussion. Governments were compelled to reactivate crisis response plans, commit to cooperation where possible, and stimulate consumption using fiscal and monetary policy tools. Yet, questions surrounding Mythos far outnumbered answers.
European Central Bank President Christine Lagarde stated in a Tuesday interview, "The consequences could be extremely serious if this technology falls into the wrong hands," referring to Mythos's potential for disruption. This concern was echoed by numerous central bank governors, finance ministers, regulators, and senior investment executives, who are actively investigating the technology's potential risks and protective measures.
The issue entered public view earlier this month following a report that U.S. Treasury Secretary Beth Cent had convened Wall Street executives, warning that an upcoming AI model could usher in a new era of autonomous cyberattacks launched by machines. Secretary Cent said in a Wednesday interview, "I am confident there is now a consensus and a concerted effort to enhance the resilience of the financial system." Outside the United States, however, the urgency conveyed by that meeting put senior foreign officials and bankers on high alert. Some fear the technology could bypass traditional cyber defenses, exposing the financial system to incalculable threats.
Bank of England Governor Andrew Bailey remarked earlier this week at an event at Columbia University in New York, "You would think the events in the Gulf would be the latest challenge facing the world." He then added, "But you wake up and find that Anthropic may have found a way to fundamentally upend the cybersecurity landscape, and you ask yourself: what did I do in a previous life?"
The core challenge for Bailey and other financial stability experts is their limited understanding of the threats associated with Mythos. It is unclear to what extent this model represents a breakthrough compared to the cyber risks they have warned about for years. Could Mythos lead to massive theft from bank accounts, paralyze international payment systems, or create a crisis of confidence that undermines the credit foundation of the financial system? Is the risk akin to the known threats posed by quantum computing—which regulators have studied in detail—or does it require entirely new countermeasures?
These questions lingered in the corridors and meeting rooms of Washington's Willard Hotel, where the Institute of International Finance hosted meetings for senior officials and corporate executives this week. The hotel was once Abraham Lincoln's residence before his inauguration. Many questions remain unanswered. It is unclear how much technical detail or access to the Mythos technology San Francisco-based Anthropic has shared with non-U.S. officials and banks. The head of U.S. operations for a European bank said they had been promised a briefing.
Limited access to information may prevent parts of the global financial system from following the IMF's advice to consistently "stay at the forefront of cyber threat developments." Canadian Finance Minister Chrystia Freeland said in a Wednesday interview that Mythos "requires our full attention" and she hoped to discuss the topic with fellow finance ministers. "It is in our collective interest to ensure the resilience of the financial system," she stated.
A source revealed this week that technical teams at the U.S. Treasury Department are seeking access to Anthropic's Mythos AI model to conduct vulnerability assessments. New York Fed President John Williams said on Thursday that recent advances in AI "demonstrate the power of these tools, both for identifying vulnerabilities and for exploiting them, at a pace that has exceeded many people's expectations." The New York Fed is a central hub for overseeing global financial transactions and stability risks.
Traditionally, financial stability risks are coordinated by the Financial Stability Board, currently chaired by Bank of England Governor Bailey. However, regarding cyber risks, the FSB often plays a secondary role, as some member countries are reluctant to share information within this broad coalition. This has led to cybersecurity issues being primarily addressed by smaller groups like the G7.
According to two sources, during a Wednesday meeting of G7 finance ministers, participants discussed the necessity of an international regulatory framework for AI governance, agreeing that no single country can manage the risks posed by Mythos alone. One source added that, given the exponential nature of the risk and potentially severe consequences, there is support for deeper discussion, though a clear path forward is lacking.
More worryingly, this doomsday-level AI technology, combined with heightened geopolitical risks, could transform from a tool for improving financial sector efficiency into a weapon that destroys its technological infrastructure. Middle East conflicts have further heightened fears of state-sponsored cyberattacks.
Officials and bankers are acutely aware that if U.S. companies can develop such disruptive AI, nations with more hostile intentions could achieve similar breakthroughs without providing advance warning to Wall Street. Laila Khawaja, Research Director at Gavekal Technologies focusing on U.S.-China tech issues, said, "The pace of cutting-edge AI development has far outstripped the creation of regulatory guardrails. It is urgent to establish these protections before such models become deeply integrated into critical sectors like finance."
This week's search for answers also highlights the "AI divide" and fragmentation within the global financial stability system. Europe is increasingly concerned about its over-reliance on foreign technology firms. Officials privately note that a growing "go-it-alone" mentality in global financial discussions could hinder effective sharing of risk information. Several European officials stated they would use this week's meetings to urge the U.S. to share information as fully as possible.
According to Swedish Finance Minister Elisabeth Svantesson, an alertness meeting involving central bank governors and finance ministers scheduled for later Thursday would include AI as a topic. "The point I want to make is that AI itself is tremendously valuable, but uncontrolled dominant power combined with AI is dangerous," she said, adding that U.S.-Israel hostilities, uncertainty, and cyber threats are core current issues.
As the global financial community debates the safety boundaries of Mythos, Anthropic officially launched an upgraded version of its flagship model, Opus 4.7, on Thursday. The company stated that Opus 4.7 performs better in software engineering, handling complex programming tasks that previously required more human oversight, with improved instruction-following and image recognition accuracy.
Notably, Anthropic explicitly stated in its announcement that Opus 4.7's overall capabilities, including in cybersecurity, are weaker than those of Mythos. The company even experimentally implemented a "differential reduction" of its cyber offense and defense capabilities during training. Furthermore, the model includes built-in automatic detection and blocking systems to prevent requests related to prohibited or high-risk cybersecurity uses.
This deployment strategy reflects Anthropic's careful balance between cutting-edge capability and safety boundaries. The company stated that the protective experience gained from the real-world deployment of Opus 4.7 will pave the way for the eventual broad commercialization of Mythos-level models. Currently, Mythos is only available to select enterprise customers to help fortify their systems.
This move also addresses market expectations regarding Anthropic's commercial progress. Amid fierce competition with OpenAI, Anthropic is seeking a new round of funding, with its valuation potentially doubling from the previous $3.8 trillion to $8 trillion or higher. Its AI programming products are gaining strong momentum, and consumer-side interest continues to rise.