India’s Computer Emergency Response Team (CERT-In) has secured access to Anthropic’s restricted Claude Mythos artificial intelligence model for cybersecurity testing. The agency will examine whether the model can help identify weaknesses in computer systems and recommend measures to address them, according to The 420.
The assessment will begin with controlled experiments rather than deployment across operational networks. Officials will study how the model performs on cybersecurity tasks, how reliably it detects vulnerabilities and what safeguards may be required before any wider use is considered.
Testing will take place in an isolated environment
CERT-In is assessing Claude Mythos in a controlled setting without access to external digital systems. The arrangement is intended to prevent the model from interacting directly with live infrastructure while the agency evaluates its capabilities and possible risks.
The testing programme will introduce tasks gradually. This approach is designed to help officials understand the model’s behaviour, determine where it can support existing cybersecurity work and identify safeguards that would be necessary for future applications.
- Purpose: assess vulnerability detection and recommendations for fixing weaknesses
- Environment: controlled testing without access to external digital systems
- Deployment approach: limited experiments before any consideration of broader use
Why access to the model is limited
Claude Mythos is available only to a limited group of trusted organisations because advanced cybersecurity capabilities can have both defensive and offensive uses. A system that assists security teams in finding weaknesses could also be misused to discover weaknesses in digital infrastructure.
That dual-use concern is why CERT-In is proceeding cautiously. The agency will assess not only whether the model can identify vulnerabilities, but also whether its use could create additional security risks. The model will not initially be connected to live external systems, and the tests will be expanded only gradually.
CERT-In is also using other artificial intelligence models for vulnerability detection. Comparing their performance with Claude Mythos could help the agency understand where advanced AI tools are useful in cybersecurity operations and where human review and additional controls remain necessary.
What the evaluation could mean for India
For Indian organisations, the exercise reflects the growing interest in using AI to support defensive cybersecurity work. If the trials show that the model can reliably identify weaknesses and suggest suitable corrective measures, such systems could eventually assist security teams with vulnerability assessment.
The immediate step is therefore an evaluation, not a nationwide deployment of Claude Mythos. CERT-In will first determine the model’s practical capabilities and risks before deciding whether it should be used more widely for cybersecurity purposes.
