Policy · Government, Defense & Space

Centre taps Sarvam AI for cybersecurity models

By Startup Enthusiast · 14 July 2026

What they do

Sarvam AI offers text-to-speech in 11 Indic languages, speech-to-text in 12 Indic languages, and translation across 23 languages

The Centre asked Sarvam and BharatGen to build advanced cybersecurity AI models to cut foreign dependence

The models will be hosted on government's isolated compute infrastructure to protect critical infrastructure

The Centre asked Sarvam AI and BharatGen to build advanced cybersecurity AI models similar to Anthropic's Mythos.
The goal is to cut dependence on foreign frontier models for safeguarding critical infrastructure.
The models will be hosted on the government's isolated compute infrastructure and deployed to protect critical infrastructure from cyberattacks.
No timeline has been set for developing these capabilities.
The government currently uses open-source and indigenous AI models, including Sarvam, to find and fix security gaps in critical infrastructure.
CERT-In is testing alternative AI models in a sandbox environment to spot vulnerabilities in critical infrastructure.
MeitY secretary S Krishnan said CERT-In's model combo currently delivers about 60-70% of Mythos’ capabilities, used to probe code, identify security gaps, and fix vulnerabilities.
Getting access to Mythos and similar advanced frontier models is a very high priority for the government, taken up with US counterparts and respective companies.

India is in talks with Anthropic and the US for access to the Mythos model.
Anthropic's Project Glasswing expanded to about 150 organisations across over 15 countries including India last month.
Wider access for Indian banks and public sector entities remains uncertain.
A MeitY department asked central ministries to hold off deploying OpenAI and Anthropic models for cybersecurity, warning against premature use.
The average time to exploit an unknown vulnerability dropped to 44 days in FY26 from 745 days in FY22, while cyberattacks in India rose to 2.9 million from 1.4 million over the same period.

Sarvam is India's full-stack sovereign AI platform, built on sovereign compute, delivering population-scale impact.
It offers voice agents (text-to-speech, speech-to-text), document digitisation, translation across 23 languages, and work agents.
Sarvam's APIs include text-to-speech in 11 Indic languages, speech-to-text in 12 Indic languages, translation across 23 languages, and document digitisation from PDFs and images.
It serves government services reaching 800 million citizens and enterprises transforming customer experience.
Sarvam claims to be sovereign by design: built, deployed, and run with full control, developed and operated entirely in India.
It offers forward deployed engineers working alongside customer teams to deliver production-ready agents.
Sarvam infrastructure has served 10B+ tokens, median latency <100ms, and uptime SLA 99.9%.
Deployment options include Sarvam Cloud (fully managed), Private Cloud/VPC, and On-Premises (air-gapped for regulated industries).
Security certifications: SOC 2 Type II, ISO 27001, DPDP compliant, role-based access, full audit trail, data residency controls.

Product: Foundation models and agentic AI systems with voice capabilities and enterprise applications
Customers: Enterprise customers across finance, aviation, telecom and government sectors
How it makes money: SaaS and licensing model for AI models and enterprise AI applications
What sets it apart: India-built sovereign AI stack with focus on cost-effective deployment and localization