Fed's Warsh Seeks Access to Anthropic's Mythos as 200-Plus Organizations Already Got In
Updated
Updated · CNBC · Jul 21
Fed's Warsh Seeks Access to Anthropic's Mythos as 200-Plus Organizations Already Got In
3 articles · Updated · CNBC · Jul 21
Summary
Kevin Warsh told the Senate last week the Federal Reserve was still trying to secure access to Anthropic's Mythos and other advanced AI models as of July 15, despite viewing them as critical for patching cyber vulnerabilities.
April's alarm came after the Fed and Treasury warned top bank CEOs that Claude Mythos Preview could expose software weaknesses; Anthropic then gave select institutions access through Project Glasswing, but not the Fed.
Roughly 50 organizations initially received Mythos in early April, and Anthropic expanded Glasswing by more than 150 organizations across 15 countries in June, leaving the central bank behind peers and some private firms.
Mythos access was further muddled by June export controls that briefly forced Anthropic to disable Mythos 5 and Fable 5 before the restrictions were eased for trusted partners and later lifted entirely.
The episode underscores broader disorder in U.S. AI oversight under Trump's more active approach, even as officials warn Chinese models such as Moonshot AI's Kimi K3 are narrowing the gap.
As China's cheaper open-source AI gains traction, can America's expensive, closed-model strategy maintain its technological and security lead?
With AI finding flaws faster than humans can patch them, are we entering an era of permanent, unmanageable cybersecurity risk?
The AI Cybersecurity Arms Race: Inside Project Glasswing, Claude Mythos, and the Global Response to Unprecedented Threats
Overview
The cybersecurity landscape is evolving rapidly as powerful AI technologies like Anthropic’s Claude Mythos Preview and OpenAI’s GPT-5.5 emerge. These frontier AI models are improving quickly, raising concerns among financial institutions and prompting urgent warnings about their capabilities. The new AI tools can empower both attackers and defenders, making cybersecurity a high-stakes domain. In response, projects like Glasswing have launched to help organizations adapt, while experts stress the need for faster, more agile security measures. This shift highlights the growing importance of advanced AI in both identifying and addressing vulnerabilities across critical systems.