In a major expansion of its global cybersecurity strategy, artificial intelligence pioneer Anthropic has announced that it is widening access to its highly restricted Claude Mythos Preview model to 15 countries, explicitly including India. The rollout, orchestrated under an industry initiative known as Project Glasswing, will distribute the elite model to an additional 150 organizations globally. This scales the program’s total cohort to roughly 200 participants, which now includes international defense bodies like NATO and the European Union Agency for Cybersecurity (ENISA).
A Model Deemed Too Powerful for the Public
First launched in April, Claude Mythos Preview boasts a massive 1-million-token context window and general reasoning capabilities that substantially surpass Anthropic’s current commercial flagship, Claude 4.7 Opus. However, the model’s true differentiator is its unprecedented aptitude for multi-step cybersecurity tasks, software engineering, and code-cracking.
During private testing, Mythos successfully discovered a 27-year-old unpatched bug in OpenBSD and repeatedly engineered active exploits for Mozilla’s Firefox JS engine. Because the model can not only identify hidden “zero-day” flaws but also analyze how bad actors might weaponize and link multiple vulnerabilities together into an attack chain, Anthropic has barred it from public release to prevent severe geopolitical misuse.
Securing Critical Infrastructure and Global Codebases
Project Glasswing enforces strict regulatory guardrails, limiting the model’s application exclusively to defensive cybersecurity operations and vulnerability patching. The newly expanded cohort shifts focus heavily toward critical infrastructure providers whose digital disruption could compromise the safety of over 100 million people. The program now spans vital, high-risk sectors including:
-
Utilities & Telecoms: Power grids, municipal water management, and nationwide communication networks.
-
Global Finance & Big Tech: Financial infrastructure giants like Euroclear, Swift, and the Intercontinental Exchange (ICE) alongside tech firms like Okta, Samsung, and SK Hynix.
-
Open-Source Maintainers: Nonprofits and developers managing foundational codebases relied upon by governments globally.
To date, the controlled deployment has yielded astonishing results, with the model accurately flagging more than 23,000 software vulnerabilities—over a quarter of which were verified as critical or severe threat vectors.
Why India’s Inclusion Signifies a Global Shift
India’s induction into Project Glasswing is particularly noteworthy because the nation is not a formal member of Western intelligence networks like NATO or the Five Eyes alliance (comprising the US, UK, Canada, Australia, and New Zealand), which form the core of the initiative.
With India rapidly scaling its digital public infrastructure, telecommunications, and financial cloud networks, Indian banks, government departments, and IT enterprises are facing an unprecedented surge in sophisticated, automated cyber threats. Gaining selective access to Mythos allows Indian defense and technology teams to actively stress-test critical software infrastructure and patch vulnerabilities before they can be exploited by malicious actors.
Anthropic estimates that rival AI organizations will likely develop and release similarly potent, “Mythos-class” cybersecurity models within the next six to twelve months. By proactively expanding Project Glasswing to key democratic hubs like India, Japan, and South Korea, the company aims to establish a robust, AI-driven global defense network before unmonitored frontier models are leaked onto the open web.
