Inside Project Glasswing: How Anthropic's AI Cyber Defense Grew to 200 Partners in 2 Months

Each new partner must meet Anthropic's security requirements before gaining access to Claude Mythos Preview and, now, Claude Mythos 5 . The company uses the model to scan partner codebases for vulnerabilities, with the explicit goal of finding and fixing critical flaws before they can be exploited by increasingly capable AI in the wild .

Important caveat: While "200+ organizations" is widely reported, Anthropic's own announcement uses the phrasing "approximately 200" and "roughly 150 new organizations." The exact number may include future commitments or rounding .

Visa and Fifth Third Bancorp join the financial cohort

On June 10, 2026, both Visa and Fifth Third Bancorp publicly confirmed their participation, deepening the initiative's presence across the global payments ecosystem .

Visa described its involvement as a proactive move to test advanced AI against its heavily fortified network. In a joint statement, President of Technology Rajat Taneja and CISO Subra Kumaraswamy said the initiative allows Visa to "test those defenses at a level that mirrors the offensive capabilities of tomorrow's most advanced threats" . Visa's network relies on zero-trust architecture, layered defenses, and automated security operations, and Glasswing gives the company a way to stress-test them against a frontier AI model.

Fifth Third Bancorp's invitation, according to CFO Bryan Preston, reflects the bank's significant role in the payments ecosystem — specifically through its Direct Express business and payroll processing capabilities . The bank joins a growing list of financial institutions that includes initial Glasswing launch partner JPMorganChase, Intercontinental Exchange (ICE), which joined on June 3, 2026, and other major U.S. banks such as Bank of America, Goldman Sachs, Citigroup, and Morgan Stanley, which have been testing the model internally, according to reports .

Claude Mythos 5: the restricted model powering Glasswing

On June 9–10, 2026, Anthropic upgraded the Glasswing program with Claude Mythos 5, the next iteration of the earlier Mythos Preview model released in April . According to AWS model documentation, Mythos 5 is Anthropic's most capable model for cybersecurity and life sciences, covering vulnerability discovery, drug design, and biodefense screening .

Key points about Mythos 5:

• It is built on the same core technology as Claude Fable 5, the publicly available version Anthropic released simultaneously — but with far fewer restrictions for Glasswing partners .
• Access is tightly controlled due to the dual-use nature of the technology. Only vetted Project Glasswing partners, government agencies, and select private companies can use it .
• The model has a 1-million-token context window and a maximum output of 128,000 tokens, enabling deep analysis of large codebases .

The rationale for the restriction is straightforward: the same model that can find vulnerabilities can also be used to exploit them. Anthropic and U.S. regulators, including Senator Mark Warner (D-VA), have argued that giving defenders a time-limited head start is the safest way to deploy such capable systems without arming adversaries .

More than 10,000 critical vulnerabilities surfaced in the first month

A May 20, 2026 update from Anthropic revealed the scale of the problem. During the first month of testing, the initial cohort of roughly 50 Glasswing partners used Claude Mythos Preview to find more than ten thousand high- or critical-severity vulnerabilities across the most systemically important software in the world .

These vulnerabilities spanned all major operating systems and critical infrastructure codebases . The model's ability to find zero-day vulnerabilities — flaws with no existing patches or signatures — revealed the fundamental weakness of traditional, signature-based defenses, which can only flag known threats . Anthropic also reported using Mythos Preview internally to scan more than 1,000 open-source projects that underpin much of the internet, including the company's own infrastructure .

Why signature-based defenses are no longer enough

The urgency driving Project Glasswing's rapid expansion is the obsolescence of traditional cybersecurity methods. Signature-based detection relies on databases of known threat patterns, but AI models like Mythos can discover entirely novel vulnerabilities that have never been seen — and, critically, demonstrate how to exploit them .

Anthropic's own testing found that Mythos Preview could "identify and then exploit zero-day vulnerabilities in every major operating system and every major web browser" . That capability, if left in the open, could be weaponized at a scale that signature-based tools cannot match. The defensive response is to give vetted defenders early access to the model so they can patch before the same capabilities reach attackers .

The initiative has also created competitive friction. When Anthropic expanded Glasswing in early June, JPMorganChase was the only bank publicly named among U.S. financial institutions receiving access . UK banks were excluded, prompting HSBC, Lloyds Banking Group, and Nationwide to accept access to OpenAI's GPT-5.5 Cyber as an alternative, according to reports . Anthropic has since signaled plans to extend Mythos access to European and UK financial institutions through its Glasswing framework, but the June 2 expansion temporarily left them out .

Project Glasswing at a glance

• April 7, 2026: Glasswing launches with 12 named launch partners and more than 40 additional organizations. Claude Mythos Preview debuts as the restricted-access model for defenders .
• May 20, 2026: Anthropic reports that early partners have found more than 10,000 high- or critical-severity vulnerabilities in the first month .
• June 2, 2026: Expansion to approximately 200 total partners across more than 15 countries. New sectors include energy, water, healthcare, and telecommunications .
• June 3, 2026: Intercontinental Exchange (ICE) announces it has deployed Mythos Preview across all businesses, including the NYSE .
• June 9–10, 2026: Claude Mythos 5 and the publicly available Claude Fable 5 are released. Visa and Fifth Third Bancorp announce their participation .

Project Glasswing is accelerating the collision between frontier AI and critical infrastructure defense. The numbers — 200 partners, 15+ countries, 10,000+ critical vulnerabilities — reflect a program designed not just to find bugs, but to reshape how the industry defends them before the offense catches up.