CERT-In Tests Anthropic Mythos AI Model for Cyber Defense

India’s national cybersecurity agency, CERT-In, has secured access to Anthropic’s advanced Mythos AI model. The Indian Computer Emergency Response Team operates under the Ministry of Electronics and Information Technology (MeitY). The agency will conduct controlled experiments with the powerful AI tool to evaluate software vulnerabilities and reinforce national digital defenses.

Government officials plan to test the technology through a cautious, phased rollout. Rather than integrating Mythos into live national security networks immediately, CERT-In will route selected non-sensitive technical tasks through the model. This step-by-step approach ensures safety while allowing experts to measure the model's actual diagnostic power.

Why Is CERT-In Testing Anthropic's Mythos Model?

Anthropic developed Mythos specifically to identify complex software security flaws and hidden system vulnerabilities. Traditional vulnerability scanners often miss subtle coding errors, but Mythos can analyze massive software codebases rapidly to spot critical weaknesses.

Key capabilities of the Mythos model include:

• Zero-Day Vulnerability Detection: Uncovering previously unknown software bugs before hackers exploit them.
• Rapid Code Analysis: Reviewing complex operating system code and browser architecture in real time.
• Preemptive Threat Mitigation: Helping developers patch software flaws before cybercriminals launch ransomware or data theft attacks.

Access to Mythos gives CERT-In an advanced diagnostic tool to audit domestic software and safeguard critical public infrastructure.

Why the Indian Government Maintained a Cautious Approach

While Mythos provides powerful defensive capabilities, the technology also carries inherent cybersecurity risks. An AI model capable of discovering software vulnerabilities for defensive patching can also generate offensive exploit code if misused.

Data privacy presents another critical concern. Because AI models operate on cloud infrastructure, sharing proprietary government code or sensitive network telemetry with an external platform poses data exposure risks. Consequently, CERT-In is conducting initial tests using limited datasets within strictly isolated sandboxes.

What Is Anthropic's Project Glasswing?

Mythos operates under Anthropic's defensive cyber initiative known as Project Glasswing. Anthropic created Project Glasswing to share advanced security AI exclusively with vetted, trusted global institutions.

Project Glasswing highlights:

• Restricted Access: Anthropic limits access to prevent bad actors from misusing the model for malicious attacks.
• Defensive Focus: The tool helps engineers audit operating systems, web browsers, and cloud networks.
• Proven Flaw Discovery: Early tests demonstrated the model's ability to locate thousands of severe software vulnerabilities across major tech platforms.

By participating in this initiative, India joins a select group of cyber authorities evaluating AI-driven defensive infrastructure.

India Moves Toward Sector-Specific AI Regulations

CERT-In's trial of Mythos coincides with India's broader efforts to establish comprehensive artificial intelligence governance frameworks.

The Telecom Regulatory Authority of India (TRAI) recently released draft guidelines for AI adoption across telecommunications. The TRAI framework mandates board-approved AI safety policies, dedicated internal oversight committees, and immediate incident-reporting mechanisms.

Simultaneously, MeitY is preparing a comprehensive public consultation paper covering AI safety, deepfake mitigation, and human-centric governance. Government policy favors sector-specific regulations, allowing individual authorities in banking, energy, and telecom to enforce tailored safeguards for critical infrastructure.

Key Overview: CERT-In and Project Glasswing Trial