Sridhar Vembu warns companies not to trust frontier AI labs, asks them to build their own AI

Zoho chief scientist Sridhar Vembu has a warning for companies with in-house IP and know-how. He has suggested that companies should be cautious with their internal data, such as critical know-how, intellectual property, and trade secrets, with frontier AI labs. He also suggested that companies should develop their own AI, saying, "Your company's know-how belongs in your own AI model," and adding that such models are only getting cheaper to train and run.

To make his point, Vembu cited Fields Medalist Terence Tao, whom he described as one of the most gifted mathematicians alive who recently had some things to say about OpenAI’s claim its AI solved the Navier-Stokes problem. Vembu quoted Tao, "The incentives may now be pointing in the direction of no longer sharing any promising research directions with the broader community, which would reverse centuries of traditions of open science and do serious long-term damage to the future of the field". In simpler terms, Tao is worried AI could extract valuable knowledge very quickly, and this could discourage mathematicians from publicly sharing promising research ideas because AI systems might immediately exploit them.

Vembu pointed to pharmaceutical companies, many of which have already built their own AI systems, often on top of open-source models, specifically so that new drug molecules and processes never touch a big AI company's servers. That caution, he wrote, is well warranted, and it isn't limited to pharma. Any company whose real value sits in a formula, a codebase, or a process nobody else has figured out faces the same exposure the moment that information gets typed into someone else's chatbot.

Earlier this week, OpenAI said an unreleased model had cracked the Navier-Stokes problem, a $1-million Clay Millennium Prize question. Mathematician Tristan Buckmaster promptly accused the company of finishing off work he was already quietly pursuing, based on hints of his approach that had leaked online. Terence Tao, watching the fallout, warned on Mastodon that incentives may now push researchers toward no longer sharing any promising research directions with the broader community, reversing centuries of open science.

Though Tao and Vembu are talking about different worlds, one is about pure mathematics and another about corporate R&D, but the underlying fear lines up. Once a frontier AI lab can see what you're working on, it can act on that information faster than you can, whether that means racing to solve your problem first or quietly absorbing it into a future model. The safest move, on this logic, is to keep the labs blind in the first place, is what Vembu is probably hinting at. - EndsPublished By: Kazi NasirPublished On: Sep 11, 2026 12:22 ISTAlso Read | Bioweapons, surveillance, espionage: Anthropic reveals how countries & groups are misusing ClaudeAlso Read | OpenAI safety researcher says 70% chance AI makes humans extinct in 3 yearsAlso Read | If you want to buy iPhone, best time to buy it was yesterday, next best is today