Anthropic triples Project Glasswing to 150 enterprise partners for Mythos AI security testing, days after a $65B raise and a confidential SEC filing.
Anthropic tripled the number of companies with access to its most powerful and most scrutinized AI models on Tuesday, a move that landed just 24 hours after the company submitted a confidential IPO filing to the SEC.
Project Glasswing, the company's controlled rollout program for Mythos-class models, now covers roughly 150 enterprise partners, up from 50. The expansion is deliberate. Anthropic is inviting select organizations to run their own codebases against a model capable of surfacing software vulnerabilities that conventional security tools miss.
That capability has been the source of friction inside Anthropic itself. The company previously flagged concerns that Mythos could expose cybersecurity weaknesses at scale, a risk that cuts both ways. The answer, for now, is controlled distribution: give the tool to enterprises with real infrastructure on the line, not the open market.
The cybersecurity angle
For companies in the program, the value proposition is direct. Organizations stress-test their own software against Mythos and get a precise accounting of where vulnerabilities sit. That is a more concrete return than most artificial intelligence deployments, which typically require months of integration before producing measurable output. Whether 150 partners is enough to validate safety at this capability level, or merely enough to look responsible before an IPO roadshow, is a legitimate question.
Yahoo Finance reported the Glasswing expansion Tuesday morning, noting that Mythos-class models are expected to reach a broader commercial release in coming weeks. Anthropic also launched Claude Opus 4.8 last week, a signal that the company is pushing its release cadence as it approaches public markets.
The IPO backdrop
On Monday, Anthropic disclosed it had submitted a draft registration statement to the SEC under Rule 135 of the Securities Act, the confidential route that preserves timing flexibility. Per Music Business Worldwide, the filing arrived days after the company closed a $65 billion Series H round on May 28, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital. That round placed Anthropic's post-money valuation at $965 billion, briefly making it the most valuable private AI company in the world, ahead of OpenAI's $852 billion mark from its March raise.
Dario Amodei, who serves as CEO, and his sister Daniela Amodei, president, founded Anthropic in 2021 after leaving OpenAI, where Dario had been vice president of research. The company structured itself as a public benefit corporation with a stated focus on AI safety. The distance between that origin story and a near-trillion-dollar valuation is something any S-1 prospectus will need to address.
OpenAI is running a parallel playbook. CNBC reported in March that the company is orienting aggressively toward enterprise productivity ahead of its own expected IPO, with CEO of Applications Fidji Simo framing ChatGPT's 900 million weekly users as a conversion funnel for high-compute business contracts. Both companies are racing to show public market investors that their platforms generate durable revenue, not just adoption numbers.
Context: the enterprise calculus
Enterprise artificial intelligence contracts are larger, stickier, and easier to justify in a prospectus than consumer subscriptions. As Digital Watch Observatory has tracked in documenting OpenAI's monetization pressures, AI companies face mounting investor expectations to convert usage into predictable revenue streams. Anthropic's Glasswing framing is shrewd: the same model capability that poses risks in adversarial hands becomes a security product when deployed internally by enterprises paying for the privilege.
There is a historical precedent for this positioning. Palantir built its early government and enterprise business on data capabilities with obvious dual-use implications, then used that contracted revenue base to go public. Whether Anthropic's safety-first reputation survives the quarterly expectations that come with being a public company is the more consequential question for 2026 and beyond.
At 150 partners, Glasswing remains small enough that Anthropic retains meaningful oversight of how Mythos is used. Once the IPO proceeds and commercial pressure builds, that control becomes harder to sustain.
---
Frequently Asked Questions
What is Project Glasswing?
Anthropic's controlled access program for its Mythos-class models, allowing select enterprise partners to run the AI against their own software to identify security vulnerabilities before broader release.
Why did Anthropic flag Mythos as a risk?
The company previously expressed concern that Mythos-class models can identify cybersecurity and programming weaknesses at a level that could be misused if widely available. The limited rollout is designed to contain that exposure while the company gathers real-world data.
What is Anthropic's current valuation?
Following a $65 billion Series H round that closed May 28, Anthropic's post-money valuation stands at $965 billion, placing it above OpenAI's most recent $852 billion figure.
How does the Glasswing expansion connect to the IPO filing?
Anthropic submitted its confidential SEC registration statement on June 1, one day before the Glasswing announcement. Expanding enterprise access to its most capable models strengthens the company's revenue narrative for prospective public investors evaluating the S-1.