Vivek Raghavan warns against foreign AI reliance as India advances domestic tech

India must develop its own artificial intelligence infrastructure to safeguard digital sovereignty and avoid becoming dependent on foreign platforms, Sarvam AI co-founder Dr. Vivek Raghavan has warned.

Speaking at The Scindia School's 129th Founder's Day event in Gwalior, Raghavan cautioned that continued reliance on imported AI models risks turning the nation into a "digital colony". He urged domestic researchers and developers to focus on building self-reliant, indigenous technology tailored to local needs and languages.

His warning comes at a time when the Indian government is actively pushing for national AI capabilities, backing domestic model development and sovereign compute infrastructure to keep data within national borders and drive local enterprise.

Global competition accelerates

Overseas tech firms are simultaneously accelerating consumer-facing AI agents. Microsoft co-founder Bill Gates described Meta's new personal agent, Muse, as "the first truly great consumer AI product" during an interview on The Verge's Decoder podcast.

Gates said he spoke directly with Meta chief AI officer Alexandr Wang, who joined the company in June 2025 from Scale AI, to congratulate him on the system. Meta launched Muse on 8 September, building it on its proprietary Muse Spark model to carry out direct tasks such as booking travel, filling out forms, online shopping, and managing email via WhatsApp or a standalone application.

On Meta's third-quarter earnings call, chief executive Mark Zuckerberg stated that more than 10 million people have adopted Muse, with around a quarter using it every day. Shopping currently accounts for the largest share of user requests in the US and Canada.

Safety concerns and open models

Despite his praise for Muse's functionality, Gates struck a cautionary tone regarding broader AI risks, describing the technology as "the most dangerous weapon of all time" and predicting two decades of major economic disruption.

He raised particular concern over open-weight models after Meta released Muse Spark Open, an 816-billion-parameter model designed to run agentic coding tasks locally across dual Mac setups. Meta stated that strong built-in safeguards accompany the model, but Gates argued that high-capability systems should only run on certified platforms where safety mechanisms cannot be bypassed or stripped away.

For Indian developers and policymakers, the rapid rollout of both open and proprietary foreign agents reinforces the urgency of Raghavan's message: domestic strength in AI is essential to national security and economic independence.