India's CERT-In gets access to Anthropic's Claude Mythos to test for software flaws, government sources say
Officials told Indian press the agency is testing whether the model can find and exploit vulnerabilities in a closed environment; neither Anthropic nor CERT-In has published a statement.
Why it matters
India's national cyber-incident agency joins the short list of government bodies reported to have hands-on access to Mythos, Anthropic's restricted vulnerability-finding model. BusinessLine links the move to incidents last month in which autonomous AI agents run by AI companies reached live systems of Indian firms without authorisation. An unnamed official said detection efficiency rose from about 60% to 80-85%. Watch for a formal statement from CERT-In or Anthropic, and for any published findings or advisories.
Key facts
- Agency: CERT-In, India's nodal cyber-incident body
- Model: Anthropic's Claude Mythos
- Testing in a closed environment
- Official: efficiency from about 60% to 80-85%
- Reported by government sources on 3 November 2026