Anthropic's Mythos AI Model Raises Alarms at U.S. Treasury and Asian Regulators
Anthropic's frontier AI model Mythos is drawing scrutiny from the U.S. Treasury and Asian financial regulators over its ability to find and exploit software vulnerabilities in banking systems. A U.S. security agency is also deploying the model internally, despite Pentagon concerns.
Anthropic's most advanced AI model, Mythos, is raising alarms among financial regulators and government agencies over its ability to identify and exploit critical software vulnerabilities faster than any previous tool.
The U.S. Treasury Department is seeking access to Mythos to study how the model could be used to probe weaknesses in the financial system. Asian regulators have separately warned banks in their jurisdictions to strengthen cybersecurity defenses in response to the model's capabilities.
Security researchers say Mythos can scan complex software systems and identify exploitable flaws in a fraction of the time it would take a human team. That speed makes it a potential tool for both defenders and attackers.
The situation is complicated by the fact that a U.S. security agency is already deploying Mythos internally, even as the Pentagon has sought to restrict Anthropic from certain government contracts. The tension reflects a broader debate within the federal government about how to manage powerful AI tools that have both defensive and offensive applications.
Anthropic has not publicly commented on the specific concerns raised by regulators. The company has previously said it takes safety seriously and works with government partners to ensure its models are used responsibly.
The CIA may have produced its first intelligence report without human involvement using an AI model, according to reporting this week. The development raises questions about accuracy and accountability in AI-generated intelligence.
OpenAI, Anthropic's main competitor, introduced GPT-5.4-Cyber this month, a defensive cybersecurity model available only through a verified-access program. The gated deployment is designed to prevent the model from being used for offensive purposes.
Venture capital investment in AI reached an all-time high of $297 billion in the first quarter of 2026, with the U.S. accounting for $250 billion of that total. Most of the investment went to AI infrastructure, models, and applications.
