Anthropic is widening access to Claude Mythos, but only for a select group of organizations that work on cyber defense, critical infrastructure, and public-sector security. The move signals that advanced generative AI is being treated less as a general-purpose tool and more as a controlled asset for identifying weaknesses before they become real attacks.
India is among the newest countries included in this early expansion, alongside Canada, Australia, New Zealand, France, Germany, Italy, Switzerland, the Netherlands, Spain, Belgium, Sweden, Japan, and South Korea. Anthropic has not disclosed which organizations in India received access.
A controlled rollout for a sensitive model
Claude Mythos first appeared in a preview form in April, when access was limited to about 50 organizations, most of them based in the United States. Anthropic had previously judged the model’s coding and vulnerability-discovery capabilities to be too powerful for a broad release without safeguards.
That caution still defines the rollout. Access remains restricted to organizations that Anthropic considers trusted and directly relevant to cyber defense missions.
The model is designed to help organizations find software flaws, detect weak points in complex codebases, and strengthen cyber resilience before attackers can exploit those gaps. Anthropic’s approach places AI in a preventive role rather than a reactive one.
Project Glasswing expands in parallel
The broader access push is happening alongside a wider expansion of Project Glasswing, Anthropic’s industry collaboration effort. The program brings together technology companies, infrastructure operators, cybersecurity experts, and governments to identify and patch software vulnerabilities proactively.
Anthropic says the initiative is not focused on offensive cyber capabilities. Instead, it is built around using AI defensively to reinforce the digital systems that support essential services across countries.
The company says participants in Project Glasswing have already discovered more than 10,000 vulnerabilities classified as high or critical. Anthropic is also exploring how AI can expand security reviews for open-source software, an important area because so much of the internet depends on it while many open-source projects lack deep security testing resources.
Who is already involved
The latest wave of participants includes organizations such as Okta, Samsung, SK Hynix, SK Telecom, Euroclear, Intercontinental Exchange, and Swift. Nato and the European Union Agency for Cybersecurity, or ENISA, have also been given access.
Earlier participants included Microsoft, Apple, Oracle, JPMorgan Chase, and CrowdStrike. Together, the list shows that Anthropic is targeting groups with large-scale digital operations and high stakes for system security.
The newest additions bring the total number of participating organizations to about 200 across more than 15 countries. The expanded roster also includes organizations in financial services, cybersecurity, technology, health care, communications, electricity, water, and hardware.
For India, the preview access places selected organizations inside a growing international AI security network. It also suggests that the country is being positioned as part of the first wave of expansion beyond the United States and the United Kingdom.
Anthropic’s timing is notable as the company enters a new growth phase. It recently filed confidentially for a U.S. initial public offering, becoming the first major frontier AI company to formally begin the path toward public markets.
Source: www.indiatoday.in-
-
-
-
