White House Blocks Anthropic Mythos Rollout as Security Fears Mount

The White House is fighting Anthropic's plan to expand access to Claude Mythos, its ultra‑powerful cybersecurity AI model, according to the New York Post. Anthropic proposed giving an additional 70 companies access to Mythos, which would bring the total number of organizations with access to 120. White House officials told Anthropic they are against broadening the rollout because of security concerns.

The standoff is the latest escalation in an increasingly tense relationship between the AI company and the Trump administration. Earlier this year, the Pentagon scrapped its contract with Anthropic and threatened to blacklist the company after it refused to give the government unchecked access to its AI tools, seeking restrictions on their use for mass surveillance or weaponry.

Mythos is Anthropic's specialized cybersecurity model built on Claude Opus 4.7. It can perform penetration testing, vulnerability identification, exploit development, and malware reverse engineering at speeds that SecurityWeek describes as compressing "time‑to‑exploit to a meaningless number of minutes." Anthropic's own analysis showed that Mythos could easily exploit electric grids, power plants, and hospitals if compromised, according to the New York Post.

The model was first announced under "Project Glasswing," a plan to provide Mythos to a select group of handpicked companies including Amazon, Google, and JPMorgan. Anthropic has also struck deals with Broadcom and others to increase computing power access, addressing concerns that limited compute could restrict growth.

On April 8, the same day Anthropic announced Project Glasswing, a handful of users managed to gain unauthorized access to Mythos, Bloomberg reported. Since gaining access, those users have been using Mythos "regularly" but not for cybersecurity purposes, according to the outlet. Anthropic confirmed last week it is investigating the unauthorized access.

The breach intensified fears about what happens if Mythos were fully unleashed without adequate guardrails. TechCrunch reported that when Anthropic restricted access to Mythos, Sam Altman called the tactic "fear‑based marketing" — before OpenAI ended up doing the same thing with its competing GPT‑5.5 Cyber model.

The White House fight over Mythos is part of a broader conflict. Defense Secretary Pete Hegseth called Anthropic CEO Dario Amodei an "ideological lunatic" after the company insisted its Claude chatbot not be used for mass surveillance against Americans or in fully autonomous weapons operations. Those demands led the Pentagon in March to designate Anthropic a national security threat and move to bar its products from the military.

Despite the tensions, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and other senior officials on April 17. A White House official told the New York Post that the administration is "actively engaging with the private sector while trying to balance innovation and security."

The Mythos standoff is creating a split market for AI cybersecurity tools. Government agencies already have access to Mythos, and the Trump administration wants to expand those permissions. But for private‑sector builders, the path to accessing frontier cybersecurity AI is narrowing, not widening.

An Anthropic spokesperson told the New York Post: "Compute is not a constraint in expanding Project Glasswing and we are engaged in collaborative conversations with the government on bringing additional parties in." The company is also releasing Claude Security, a separate product now in public beta for Claude Enterprise customers, which scans codebases for vulnerabilities without the same level of access restrictions.

For builders, the key takeaway: frontier cybersecurity AI is entering a gated era. If you're building security tools, expect more access restrictions, more application processes, and more government oversight. Claude Security may be the more accessible path for most teams, while Mythos‑level capabilities remain locked behind government and enterprise gates.

Both Anthropic and OpenAI are now restricting access to their most powerful cybersecurity models. TechCrunch notes that OpenAI's GPT‑5.5 Cyber will be available only to TechCrunch reports, will be limited to so‑called critical cyber defenders who submit an application with credentials and planned use. This pattern — powerful AI behind gated access — is becoming the industry standard, not the exception.

What's different about the Mythos situation is the direct White House involvement. The government isn't just setting policy; it's actively negotiating which companies get access to specific AI models. That's a new level of public‑private control over frontier AI, and it sets a precedent that will shape how every future high‑capability model gets deployed.