Anthropic expanded Project Glasswing, its initiative to find and patch critical software vulnerabilities with AI, to roughly 150 new organizations across more than 15 countries on Tuesday. The program puts Claude Mythos — Anthropic's most capable model to date — inside the codebases of power, water, healthcare, communications, and hardware operators. The expansion lands one day after Anthropic filed confidentially for an IPO, following a $65 billion funding round at a nearly $1 trillion valuation.
Claude Mythos is the engine. Anthropic says the model can surface thousands of zero-day vulnerabilities over several weeks of scanning, a throughput that would be impossible for human security teams at the same cost. In early April, Anthropic seeded 50 initial partners, including the U.S. government, with Mythos Preview access to test the model against production code.
The new cohort fills sector gaps Anthropic flagged in the first wave. Power utilities, water systems, hospitals, telecoms, and chip and hardware vendors were under-represented in April, and many of the additions maintain open-source or shared codebases that other organizations and governments depend on. The Financial Times reported the expanded list spans Australia, Canada, France, Germany, Italy, Switzerland, the Netherlands, Spain, Belgium, Sweden, India, Japan, New Zealand, and South Korea.
Key facts
- 01Anthropic expanded Project Glasswing to roughly 150 new organizations across more than 15 countries on Tuesday.
- 02Claude Mythos can identify thousands of zero-day vulnerabilities over several weeks, per Anthropic.
- 03The initial April cohort included 50 partners, among them the U.S. government, scanning code with Mythos Preview.
- 04Expansion follows Anthropic's $65B funding round at a nearly $1 trillion valuation and a confidential IPO filing.
- 05New partners reportedly include Okta, Samsung, SK Hynix, SK Telecom, NATO, and EU cybersecurity agency ENISA.
Reported participants include identity provider Okta, Samsung, SK Hynix, SK Telecom, NATO, and ENISA, the European Union's cybersecurity agency headquartered alongside other EU bodies. Anthropic has not confirmed the full list. The common thread, in Anthropic's words, is consequence: a breach at any one of these targets cascades into civilian and government systems downstream.
“What each partner has in common is that a successful attack on their codebase could be catastrophic.”— Anthropic, company blog post
Anthropic estimated that for most partners, a successful attack on their codebase could affect more than 100 million people, with knock-on effects for both national and global security. That framing is also the argument for why Mythos is being handed out rather than sold narrowly — Anthropic wants the defensive use case established before the same capability shows up elsewhere.
That elsewhere is already arriving. OpenAI released GPT-5.5-Cyber, a cybersecurity-tuned model, and has rolled it out to a large group of partners for testing. Anthropic has been explicit that it expects rival labs to match Mythos Preview's vulnerability-discovery capabilities in short order, which is why the company is racing to lock in Glasswing's defender-first deployment pattern now rather than later.
The strategic logic tracks with Anthropic's broader posture. The company has positioned safety-aligned deployment as a commercial moat, not just a policy stance, and Glasswing is the cleanest expression of that pitch to date: take the most dual-use capability in the model — finding software flaws nobody else can find — and put it exclusively in the hands of the people whose job is patching them. The same model run by an attacker would be a weapon; run by Okta or ENISA, it is an audit.
The risks are real and unresolved. Concentrating zero-day discovery inside a single vendor's model creates a high-value target in its own right. Disclosure timelines across 150 organizations in 15-plus jurisdictions will be operationally messy, and any leak of Mythos outputs — a list of unpatched flaws in critical infrastructure — would be precisely the catastrophic event Anthropic says it is trying to prevent. The company has not detailed how findings flow from Mythos to the affected partners, or what happens when a flaw sits in a shared dependency used by organizations outside the program.
Glasswing is shaping up to be Anthropic's most concrete bid to convert frontier model capability into infrastructure-grade trust before the IPO road show begins. If the program holds — meaning no high-profile leak, a steady cadence of patched zero-days, and visible buy-in from NATO and EU agencies — it gives Anthropic something OpenAI cannot easily replicate: a defender-side reference customer list across critical sectors. That is a useful story to tell public-market investors evaluating a near-trillion-dollar AI company whose biggest open question is durable enterprise demand.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
