Enterprise

Anthropic opens Claude Mythos to 150 critical-infrastructure operators a day after filing for IPO

Project Glasswing's expansion into power, water, healthcare and telecom lands the same week Anthropic confidentially submitted its S-1 at a near-$1 trillion valuation.

By Asher Glick · Enterprise software desk · 2026-06-08 · 4 min read

Photo: Unsplash / Matthew Henry — High-voltage transmission lines against a clear sky, representing critical power infrastructure

Anthropic on Tuesday opened early access to Claude Mythos Preview, its most capable and most restricted model, to 150 additional operators across more than 15 countries, expanding Project Glasswing into the power, water, healthcare, communications, and hardware sectors a day after the company confidentially filed its S-1 with the SEC.

The sequencing isn’t subtle. Glasswing launched in early April with 50 partners, including the U.S. government, alongside Apple, Nvidia, Microsoft, CrowdStrike and Palo Alto Networks. Tuesday’s cohort, according to the Financial Times reporting cited by TechCrunch, pulls in Okta, Rubrik, NATO, and a heavy contingent of South Korean industrial players including Samsung, SK Hynix, and SK Telecom. “What each partner has in common is that a successful attack on their codebase could be catastrophic,” Anthropic said, framing the program around operators whose systems other governments and organizations depend on, and where a major incident could affect more than 100 million people per partner.

That’s the pitch. It also happens to be the IPO narrative.

Anthropic’s confidential prospectus, filed Monday, lands against a $965 billion private valuation, a $47 billion run rate disclosed in May, and $10 billion in revenue last year. OpenAI was valued at $852 billion in late March. Glasswing isn’t just a safety program; it’s a distribution moat dressed in safety vocabulary, the kind of story underwriters will want printed across the roadshow deck.

The numbers underneath are real, which is what makes the framing work. Since April, Glasswing partners have surfaced more than 10,000 high- or critical-severity vulnerabilities. Cloudflare identified 2,000 bugs across its critical-path systems, 400 of them high or critical. Mozilla found and fixed 271 vulnerabilities in Firefox 150, more than ten times what a prior Anthropic model produced on an earlier version of the browser. Claude Security, built on the publicly available Claude Opus 4.8, patched more than 2,100 vulnerabilities in three weeks.

The political backdrop closed the loop the same day. The Trump administration signed a scaled-back executive order on AI security, establishing a voluntary framework under which developers may submit advanced models for government review up to 30 days before public release. Voluntary is the operative word. Glasswing is what compliance looks like when the regulator has stepped back and the incumbent gets to define the standard.

It’s, in the older sense of the phrase, regulatory capture by press release, executed the week the bankers were reading.

Sources