AI model developers need to take the lead on safety: Ashwini Vaishnaw

Union minister Ashwini Vaishnaw said that regulating artificial intelligence (AI) will take both law and technical safeguards, and that companies building the technology must take the lead on safety.

In an interview with ET, when asked about the debate in the US over slowing or pacing the technology, Vaishnaw said governments have stepped in before when disruptive technologies arrived.

"Whenever any disruptive technology has come into the world — whether automobiles or the nuclear industry — governments and parliaments have stepped in to have the right kind of regulation. We have to work in a techno-legal framework. It cannot be just a law that solves this problem. It has to be a technical as well as a legal solution," he said.


Vaishnaw's comments come in the backdrop of the AI safety debate triggered by incidents involving AI agents. A number of OpenAI's autonomous agents went rogue and attacked the model-hosting platform Hugging Face, and OpenAI didn't know until Hugging Face told it several weeks later.

The ChatGPT maker has since disclosed six new incidents in which agents made up data, moved files onto the public internet without permission, and hid their mistakes from humans. Anthropic CEO Dario Amodei then publicly called for a pause on new frontier models so the industry could build proper safety rails, with backing from OpenAI CEO Sam Altman and Elon Musk.

Asked whether there is a greater responsibility on AI model developers for safety, Vaishnaw said yes, anyone creating a technology should make sure it is safe.

"What are the liabilities of agents? Who is liable if an agent does something very serious, such as breaking down a grid or pushing a student towards suicide? Or allowing or forcing a car to suddenly brake in the middle of traffic? Those questions have to be answered. The industry has to answer them and take the lead because, unlike physical technology, where it is easier to find what to regulate and what to check, that kind of measurement in AI is very difficult," he said.

Vaishnaw added that the people who design and build the models know where the problems lie, and that they need to be part of the regulatory process.

The union minister also spoke about the India Semiconductor Mission 2.0, saying that the next wave of semiconductor investments are expected to begin within six months, as the government looks to build on nearly $23.5 billion committed under the first two phases of its chip mission.