India Launches Indigenous AI Models; Security Concerns Rise at OpenAI

The Government of India has taken a significant step in the field of Artificial Intelligence by launching indigenous foundation models. Under the India AI Mission of the Ministry of Electronics and Information Technology (MeitY), the EKA-1, Dhir 2B, and Dhir 5B models have been introduced. According to a report by The Hindu, the EKA-1 model was developed by Socket Labs Technology and Research, while the Dhir 2B and Dhir 5B models were prepared by Zentek AItech Innovations. The models were unveiled in the presence of IT Secretary S. Krishnan and senior officials of the ministry.

The EKA-1 model has been specifically designed for Indian languages, banking, and cyber security. It possesses 26.3 billion parameters and has been trained on approximately 7.5 trillion tokens. This model covers all 22 official Indian languages, including English. According to Abhishek Upravle, CEO of Socket, this model has been tailored to suit India's banking, financial services, and cyber security landscape. Notably, EKA-1 can operate on local devices such as smartphones even without mobile data or Wi-Fi.

On the other hand, the Dhir 2B and Dhir 5B models developed by Zentek are open-source models that support English and 10 Indian languages. These models can function in sensitive areas without online connectivity across the defence, government, industry, and education sectors. Prof. Shashikumar, CEO of Zentek, stated that these models can run on various computing platforms such as GPUs and TPUs, ensuring that data remains on the device, thereby enhancing privacy and security. IT Secretary S. Krishnan believes that such indigenous models will strengthen India's technological and strategic autonomy.

While India is enhancing its capabilities, companies like OpenAI and Anthropic are facing security challenges globally. According to a report by The Times of India, OpenAI and Anthropic have instructed their employees to hide company identification badges and logos. These steps have been taken due to rising hostility towards AI employees and fears of espionage. Recently, there have been instances where confidential information was sought from some employees, and security concerns have increased due to attacks and protests against executives.

According to an NDTV report, warnings regarding flaws in operational security at OpenAI had been issued for years. Even before the launch of ChatGPT in 2022, some employees had informed the management and the board that security protocols were too lax, potentially allowing foreign governments to easily obtain sensitive information. According to employees, the management was focusing more on competing with rivals like Google and Anthropic than on security. Will Dipu, a former researcher who resigned in April, stated that AI leaders cannot sacrifice speed for security.

Regarding this matter, OpenAI spokesperson Alana Widman stated that OpenAI is one of the most targeted organisations in the world; therefore, the company invests heavily in security programmes to protect against insiders and cybercriminals. Currently, Thibault Sotioux, Senior Executive of Product Development, is leading the initiative to secure internal systems. In August, he redeployed a quarter of the product engineering team to improve security infrastructure.

Meanwhile, according to a report by Business Standard, the Q2 FY27 report of Quess Corp reveals that despite a 3.5 per cent decline in GCC tech hiring, demand for individuals with AI, data, and cyber security skills has increased. Such specialised skills are driving 10 out of 3 tech hiring demands, with an increasing contribution from Tier-2 cities.