India Tests Systems Against Mythos Cyber Threat

India is testing financial and government digital systems, including the Aadhaar national identity system and government login platforms, against vulnerabilities linked to Anthropic’s next-generation AI model, Mythos. The tests involve Indian technology firms, the national cybersecurity agency CERT-In, and critical digital infrastructure used by public institutions and banks.

Mythos and Frontier AI Security

Anthropic unveiled Claude Mythos Preview on 7 April 2026. The model has been associated with autonomous discovery and exploitation of zero-day vulnerabilities in operating systems and web browsers. Zero-day vulnerabilities are software flaws that are unknown to the vendor or remain unpatched at the time of discovery.

Indian Testing Framework

Infosys and Tata Consultancy Services are participating in the tests in secure environments. They are using Anthropic’s Claude Opus 4.7 model because they do not yet have direct access to Mythos. CERT-In is conducting tests on critical digital infrastructure, and the Reserve Bank of India held meetings with domestic banks in May 2026 on possible risks linked to Mythos.

Project Glasswing and Access Controls

Anthropic launched Project Glasswing with Mythos Preview and granted limited access to about 50 organisations and governments. The participants include the UK government, Mozilla, Palo Alto Networks, and Cloudflare. Cloudflare has observed the model’s ability to construct exploit chains and generate proofs of concept for identified bugs.

Important Facts for Exams

• CERT-In is India’s national cybersecurity agency under the Ministry of Electronics and Information Technology.
• Aadhaar is India’s 12-digit unique identity system managed by the Unique Identification Authority of India.
• NIST stands for the National Institute of Standards and Technology, a United States federal agency.
• Zero-day vulnerabilities are a major category of cyber risk in operating systems, browsers, and enterprise software.

Vulnerability Findings and Global Response

On 25 May 2026, Anthropic said Mythos Preview had identified over 23,000 potential vulnerabilities across more than 1,000 open-source software projects. Of these, 1,726 were confirmed, including more than 1,000 rated high or critical severity.

United States AI Security Measures

On 6 May 2026, NIST announced that its Center for AI Standards and Innovation will conduct pre-deployment evaluations of frontier AI models. The evaluations cover models from Google, Microsoft, and xAI and focus on cybersecurity risks.