Anthropic Adds 150 Partners to Claude Mythos Program

Anthropic is expanding Project Glasswing to give roughly 150 more organizations controlled access to its Claude Mythos model while the company has confidentially submitted a draft S‑1 registration statement to the U.S. Securities and Exchange Commission as a step toward a possible initial public offering.

The new cohort includes companies and agencies that maintain widely used software in power, water, healthcare, communications and hardware. Each organization must meet Anthropic’s security requirements before gaining access, the company wrote, and many partners support systems used by governments and hundreds of millions of people.

Anthropic described the expansion as a defensive measure. Participants who used a Mythos preview have identified more than 10,000 high- or critical-severity software vulnerabilities, according to the company. Partners have used the model to write patches, run security reviews and test software releases. Anthropic added it expects to make Mythos-class models available more broadly “in the coming weeks” after additional safeguards are implemented.

Draft Anthropic materials leaked in March, and subsequent testing and reviews raised questions about the model’s capabilities. Independent testing found the model could autonomously complete a multi-step simulated corporate network attack. Engineers reported the model helped identify hundreds of issues in widely used software, and security researchers said it assisted in developing exploit chains for recent chipsets.

Anthropic warned that other developers may release models with comparable capabilities within six to 12 months and that some releases could lack similar safeguards. The company wrote that expanding access for defenders aims to provide stronger defensive tools before attackers gain access to equivalent systems. “If we’re successful, we hope to enable a permanent advantage for defenders,” Anthropic wrote.

The confidential S‑1 filing allows Anthropic to pursue an IPO after the SEC completes its review and market conditions permit. The submission does not commit the company to a specific timeline or offering size.

Anthropic said the controlled rollout has produced practical results for partners, including faster discovery of vulnerabilities and accelerated remediation work. The company estimated a major attack on most partners could affect more than 100 million people and have significant implications for global and national security.

Anthropic also announced a new Opus 4.8 model and repeated that wider customer access to Mythos-class models depends on completing technical safeguards, partner reviews and compliance checks under Project Glasswing.

Regulators, security researchers and some governments have been monitoring Mythos. The model’s ability to support multi-step exploit development and surface overlooked vulnerabilities has prompted calls for controlled testing, oversight and coordination between model developers and critical infrastructure operators. Anthropic’s expanded program aims to increase that coordination while the company prepares the model for broader commercial use.