US Federal Reserve Warns of Advanced AI Model Risks Amid Access Delays
The US Federal Reserve has warned about the cybersecurity risks posed by Claude Mythos Preview, an advanced artificial intelligence model developed by Anthropic, despite being unable to access the model itself for several months. The situation has raised concerns about the central bank’s ability to keep pace with rapidly evolving cyber threats.
According to reports, the Federal Reserve and the US Department of the Treasury held a special meeting with senior banking executives in April to warn them about the model’s advanced capabilities, particularly its ability to identify security vulnerabilities. By that time, several financial institutions had already begun strengthening their cyber defenses using the technology.
Anthropic said that around 50 organizations were granted access to the model through its Project Glasswing initiative, including major financial institutions such as JPMorgan Chase and leading technology companies including Amazon, Apple, and Google. Meanwhile, the Federal Reserve reportedly continued seeking access until mid-July.
Speaking before the US Senate, Federal Reserve Chairman Kevin Warsh said the central bank does not control access to the model, emphasizing that financial institutions should be able to use advanced AI systems to better protect themselves against increasingly sophisticated cyber threats.
In June, Anthropic temporarily disabled the updated Mythos 5 model to comply with US national security directives. The model was later reactivated for a limited number of approved partners following authorization from the US Department of Commerce.
The developments come as the administration of President Donald Trump expands its role in regulating the artificial intelligence sector, amid a series of high-profile resignations involving senior officials responsible for AI policy.
At the same time, competition has intensified after Chinese startup Moonshot AI launched its Kimi K3 model, which outperformed several US AI models in benchmark tests, fueling concerns across the technology and policy communities over China’s accelerating progress in artificial intelligence.
Former White House AI adviser David Sacks described the performance of the Chinese model as “concerning,” arguing that the United States is “tying itself in knots” with restrictive policies that could undermine its position in the global AI race.
Daniel Newman, CEO of research firm Futurum, said the Federal Reserve’s lack of access to the latest AI models leaves it “playing catch-up,” adding that major financial institutions face continuous waves of innovation from both the United States and China, placing additional pressure on the resilience of the US financial system.
