Govt prioritises access to Anthropic’s Mythos AI model to strengthen cybersecurity: IT Secretary

New Delhi, July 13 (IANS) The government is actively pursuing access to Anthropic’s advanced AI model, Mythos, as part of its efforts to bolster cybersecurity and identify software vulnerabilities, while simultaneously preparing with alternative AI models in anticipation of securing access, IT Secretary S. Krishnan said on Monday.

Speaking about the government’s efforts, Krishnan said discussions are underway with US authorities and the company to obtain access to Mythos and other frontier AI models.

“While the process is still under negotiation, the government is not waiting idly and has already begun using alternative AI models to identify and address software vulnerabilities,” he noted.

“Mythos has demonstrated an ability to uncover vulnerabilities in widely used software that often remained unresolved because they were considered low priority, too expensive to fix, or had simply gone undetected,” Krishnan said.

He said the technology presents an opportunity to systematically identify such weaknesses and correct them, making access to Mythos and similar advanced AI systems a key priority for the government.

“The government has taken up the matter with its counterparts in the United States as well as the companies developing these frontier AI models. However, access involves a formal process that is currently being negotiated,” he mentioned.

In the meantime, the Indian Computer Emergency Response Team (CERT-In) has established a dedicated sandbox, or “war room”, where alternative AI models are being used to test software code, detect vulnerabilities and develop secure workflows.

These substitute models are estimated to offer around 60 to 70 per cent of the capabilities of Mythos, enabling officials to conduct extensive testing even before gaining access to the more advanced system.

Krishnan said CERT-In is probing different elements of software code, identifying vulnerabilities and fixing them through this controlled environment. He described the ongoing exercise as a “dry run” designed to ensure that systems and processes are ready once access to Mythos becomes available.

He added that the initiative is also helping the government tackle a significant proportion of software vulnerabilities despite not yet having access to the advanced AI model.

Referring to current US export restrictions on advanced AI technologies, Krishnan acknowledged that such controls remain a hurdle in securing access to frontier AI systems.

–IANS

pk