CERT-In has gained access to Anthropic's restricted Claude Mythos model for controlled software-vulnerability tests. A senior official said the agency is using limited datasets in an isolated environment, limiting exposure of sensitive information. Anthropic has not detailed the arrangement public. CERT-In is expected to widen testing gradually if early experiments prove useful later still now.

India's cybersecurity agency CERT-In has gained access to Anthropic's Mythos AI model and will begin using its capabilities, adopting a cautious and calibrated approach to assess how best it can be utilised, IT Ministry sources said on Tuesday.

Asked how the agency is going about the access, and whether it will take a more cautious approach, the sources said CERT-In will start using and experimenting with the model.

"We will see how best we can use it," the sources said, adding that the process will be slow.

"We will slowly route anything we want to test. So it will be a slow process," they said, adding that the whole idea is that "the cure should not be worse than the problem."