What Happens When AI Gets Drug Discovery Wrong | Ryan Cawood & Raphaël Ognar

An AI fabricated a scientific citation. When Raphaël Ognar pushed back, it apologized, saying it had not wanted to fall short by giving him an unsatisfactory answer. He told it that "I don't know" would have been better than lying, and stopped using it that day. That anecdote opens a conversation that escalates. Ryan Cawood pastes in a DNA sequence he already knows the answer to and gets a confident, wrong one back. A cell line takes nine months to build and fails at the door of a GMP facility because a chain-of-custody record has a hole in it. And the FDA does not care where your data came from: submit it, and you own it. Ryan Cawood, PhD, is Co-Founder and CEO of Lab Thread. Raphaël Ognar is President, CEO, Co-Founder and Chairman of NKILT Therapeutics, and carried two small molecules through to approval inside big pharma before leaving to build his own company. Both run companies with AI in the workflow. Neither will put it in the pitch deck. (00:00) "I don't know" is a good answer(00:24) What happens when AI goes wrong in drug discovery(01:40) The publication that did not exist(03:41) The AI apologized for not wanting to disappoint(05:02) The DNA sequence it read confidently and wrong(06:51) Who is liable when the model is wrong(10:01) What a mistake actually looks like in a lab(12:06) Build checkpoints into the prompt, not just at the end(14:47) Flawed data in, confident answer out(16:29) Failure at big pharma versus failure at a small biotech(21:01) Trials are designed for approval, not for the real world(25:16) When you cannot show where the answer came from(27:25) Nine months of chain of custody, recorded on paper(30:20) Confidence is not accuracy(32:34) Integrating AI instead of renting it(35:24) Greed versus outcome in the AI bubble(37:04) Who is really paying for your AI subscription(39:33) The energy and water nobody prices in(41:28) AI for AI's sake, and the colleague who rewrites your email(43:31) The pitch deck with no mention of AI(46:07) What inspired you to take this path(50:20) The hardest decision