Anthropic recently developed Claude Mythos, a powerful AI capable of uncovering thousands of security vulnerabilities across major operating systems and web browsers, which the company now refuses to release to the public due to its dangerous capabilities. This emergent hacking prowess, which discovered a 27-year-old bug in OpenBSD, was not an intended feature but a byproduct of enhancing the AI’s intelligence. In response to unauthorized access to the model on its announcement day via contractor credentials, Anthropic initiated Project Glasswing, allowing 40 selected companies, including Apple and Microsoft, to use the model for security enhancements before potential attackers could replicate its capabilities.

Citi: Citi is a major multinational investment bank and financial services corporation providing banking, credit cards, and wealth management globally. It collaborates with regulators on cybersecurity and economic risks. In the news, it attended the emergency meeting convened by the U.S. Treasury Secretary over the Claude Mythos incident.
Apple: Apple develops hardware, software, and services centered on personal computing, mobile devices, and operating systems like iOS and macOS. As a key participant in cybersecurity initiatives, it accesses advanced tools for vulnerability management. In the news, it received access to Claude Mythos under Anthropic’s Project Glasswing to secure its software products.
Google: Google builds internet services, cloud infrastructure, and the Android operating system while advancing AI and security technologies. It partners on defensive cybersecurity efforts against emerging AI threats. In the news, it gained access to Claude Mythos via Project Glasswing to identify vulnerabilities in its major platforms.
Mozilla: Mozilla is a non-profit organization maintaining the Firefox web browser and promoting open web standards. It leverages cutting-edge tools for rapid security improvements in its projects. In the news, it used Claude Mythos to identify and patch vulnerabilities in Firefox.
JPMorgan: JPMorgan Chase is a leading U.S. bank delivering commercial, investment, and asset management services worldwide. It works closely with regulators on technology-related financial risks. In the news, it received Claude Mythos access through Project Glasswing and joined the Treasury Secretary’s emergency meeting on the breach.
Anthropic: Anthropic is an AI safety and research company that builds reliable, interpretable, and steerable AI systems including the Claude large language models. It recently introduced Claude Mythos Preview, a frontier model exhibiting advanced cybersecurity capabilities such as autonomously discovering complex software vulnerabilities, and launched Project Glasswing to provide limited access to trusted partners for defensive purposes. In the news, Anthropic withheld public release of the model due to risks of misuse after unauthorized access occurred shortly after announcement.
Microsoft: Microsoft provides operating systems like Windows, enterprise cloud services through Azure, and developer tools with a focus on secure computing. It joins industry coalitions to proactively address software flaws. In the news, it was granted access to Anthropic’s Claude Mythos under Project Glasswing for enhancing defenses.
Goldman Sachs: Goldman Sachs is a prominent global investment bank offering services in securities, investment management, and consumer banking. It engages with government officials on financial stability issues amid emerging technology risks. In the news, it participated in the U.S. Treasury Secretary’s emergency meeting addressing threats from Anthropic’s Claude Mythos.
Scott Bessent, U.S. Treasury Secretary: Scott Bessent is the current U.S. Secretary of the Treasury responsible for managing federal finances, economic policy, and financial stability. He recently announced key appointments at the Treasury Department and testified on budget matters before Congress. In the news, he called an emergency meeting with major banks like Goldman Sachs and Citi specifically in response to the Claude Mythos security breach.

`json
{
“Breach Response”: “Unauthorized access to the model occurred on the announcement day via contractor credentials, leading to immediate government involvement and accelerated software patching efforts.”,
“Project Glasswing”: “An initiative by Anthropic to provide select organizations with access to Claude Mythos Preview, aiming to address software vulnerabilities before potential misuse by malicious entities.”,
“Emergent Capabilities”: “Claude Mythos developed hacking abilities unintentionally, revealing significant vulnerabilities, including a long-undetected flaw in a major operating system.”
}
`