Headline: Anthropic widens private rollout of Claude Mythos via Project Glasswing as it quietly files for IPO
Anthropic is expanding controlled access to its powerful Claude Mythos model through Project Glasswing, a security-focused program that lets vetted tech firms, security researchers and governments probe the system for bugs and exploits before any public release. In a blog post Tuesday, Anthropic said it is adding roughly 150 organizations to the program as it prepares a broader release — once additional safeguards are completed — and after confidentially filing for an IPO with the U.S. Securities and Exchange Commission.
Project Glasswing’s latest intake focuses on organizations that operate critical infrastructure across power, water, healthcare, communications and hardware. Anthropic says many of the new partners maintain software used by governments and hundreds of millions of people, and that a successful attack on their codebases could be catastrophic — “for most partners, we estimate that a major attack could affect more than 100 million people, with important ramifications for both global and national security.” Each new partner must meet Anthropic’s security requirements before gaining access.
Anthropic says early Project Glasswing partners using Mythos Preview have already helped identify more than 10,000 high- or critical-severity software vulnerabilities. Partners have used the model to write patches, run security reviews and test code prior to release. Notable third-party results cited by Anthropic and other researchers include the U.K. AI Security Institute’s simulated 32-step corporate network attack, Mozilla’s report that Mythos identified 271 vulnerabilities it fixed in Firefox, and a claim from security startup Calif that a preview version helped researchers develop an exploit chain targeting Apple’s M5 chips.
Those capabilities are precisely why the model has drawn scrutiny from researchers and regulators: powerful AI can speed both defensive work and offensive cyber operations. Anthropic argues defenders need equally advanced tools and is moving cautiously because it expects other developers to release Mythos-class models within six to 12 months — potentially without comparable safeguards. “In the future, frontier model releases will become increasingly high-stakes,” Anthropic wrote, adding that the company hopes to “enable a permanent advantage for defenders.”
The expansion and IPO filing arrive as market sentiment shifts: users on Myriad, a prediction market run by Decrypt’s parent company Dastan, have dialed back their odds that Claude Mythos will reach a public release by the end of June — the market currently assigns a 26.5% chance, down from a 58.5% peak last week.
Why crypto readers should care: as AI speeds up vulnerability discovery and exploit development, exchanges, wallet providers, node operators and other crypto infrastructure could be affected — both at risk of AI-assisted attacks and able to benefit from AI-driven defenses. Anthropic’s controlled rollout is a test case in balancing rapid capability growth with caution.
Read more AI-generated news on: undefined/news
