Background: A Frontier Model Held Back From Public Release
<cite index="18-3">The cybersecurity community went on alert when Anthropic announced on April 7, 2026 that its latest and most capable general-purpose large language model (LLM), Claude Mythos Preview, had demonstrated remarkable — and unintended — capabilities.</cite> <cite index="21-3">Claude Mythos Preview is a general-purpose, unreleased frontier model that reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.</cite>
<cite index="18-6">Claiming that it would be too risky to release the model, and that the company had the moral responsibility to disclose these vulnerabilities, Anthropic said it would not immediately offer the model to the public.</cite> <cite index="18-7">Instead, it granted exclusive access to tech giants to test the model's capabilities, a process Anthropic dubbed Project Glasswing.</cite> <cite index="23-3">Initial Project Glasswing partners included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks.</cite>
<cite index="21-4">Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.</cite> In a notable demonstration of autonomous capability, <cite index="18-18">an evaluation by the AI Security Institute (AISI) found that Mythos was able to take over a simulated corporate network in three out of ten tries — the first AI model to succeed at the task.</cite>
US–Japan Diplomatic Channel
<cite index="4-1,4-2">Japan's government and some banks were given access to Anthropic's new Claude Mythos artificial intelligence model with advanced cybersecurity capabilities, the government said, as concerns mount about cyberattacks led by sophisticated AI.</cite> The path to access was brokered at the diplomatic level: <cite index="5-4">US Treasury Secretary Scott Bessent said during his three-day visit to Japan through May 13 that the Japanese government and the country's financial institutions would be given access within two weeks.</cite>
<cite index="5-5">Sources close to the matter said Japan's three megabanks — MUFG Bank, Sumitomo Mitsui Banking Corp and Mizuho Bank — are expected to gain access to the sophisticated AI model.</cite> <cite index="4-3">Finance Minister Satsuki Katayama told reporters that "there is no doubt that this will contribute to strengthening security" at a time when it is "unacceptable" for Japanese financial institutions to fall behind other countries in this field.</cite>
<cite index="3-6,3-7">Anthropic announced it would expand access to Mythos to about 150 organizations across more than 15 countries, with the Japanese government and several domestic financial institutions included in this expansion.</cite> <cite index="5-9,5-10">Japan's Financial Services Agency (FSA) urged financial institutions to implement short-term countermeasures in response to heightened cyberattack risks posed by sophisticated AI, and has launched a framework for public- and private-sector officials to discuss stronger responses to Claude Mythos, consulting with the Bank of Japan and major banks.</cite>
Export Controls and Government–Industry Tensions
The expansion of allied access came against a turbulent backdrop of US government intervention in Mythos's deployment. <cite index="11-16">On June 12, 2026, US Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei requiring a license for the export, re-export, or transfer of two of the company's top-tier models, Claude Mythos 5 and Claude Fable 5, citing cybersecurity capability concerns.</cite> <cite index="13-7">In order to comply with the order, Anthropic said it had to disable access to both.</cite>
The episode was resolved through negotiation. <cite index="14-2">Anthropic said it would "scale up our government collaboration" and work to develop a "shared industry framework" with the administration and the private sector.</cite> <cite index="14-10">The company said it would prerelease frontier models for federal authorities to review and test for weaknesses, and implement dedicated research teams of Anthropic employees to ensure its models would not run afoul of government priorities.</cite>
Broader Regulatory Context
The Mythos access framework fits within a wider pattern of structured government engagement with frontier AI. <cite index="7-6">Anthropic offers Claude for Government, which supports Federal Risk and Authorization Management Program (FedRAMP) High workloads that enable federal workers to use Claude for sensitive unclassified work.</cite> <cite index="8-4,8-5">In July 2025, four US AI companies entered into contracts with the Department of Defense (DoD) to accelerate adoption of advanced AI capabilities, with DoD awarding up to $200 million each to Anthropic, Google, OpenAI, and xAI.</cite>
<cite index="24-14">Anthropic plans to work with partners, including the US government and allied governments, to expand Project Glasswing, and intends to make Mythos-class models generally available, but only after developing stronger safeguards.</cite> <cite index="12-4">The age of governments ignoring capable model releases is likely over.</cite>