GB Greg Brockman Greg Brockman Co-founder and President of OpenAI. Earlier an early employee and the first CTO of Stripe, and at OpenAI has worked on projects including OpenAI Gym and OpenAI...

“As the models have got more capable, they can do more kind of within just the forward path without having to actually write out that thinking, which means we need a new approach to monitoring.”

AI Explained Official Podcast · AI Research & Frontier Labs · November 2026

“As the models have got more capable, they can do more kind of within just the forward path without having to actually write out that thinking, which means we need a new approach to monitoring.” — Greg Brockman, AI Explained Official Podcast

Context

A clip played on AI Explained, introduced as OpenAI President Greg Brockman. The host contrasts it with Brockman's earlier enthusiasm for monitoring a model's written chain of thought, which gets less informative as more reasoning happens inside the model.

Transcript

AI Explained Official Podcast Around 08:22 into the episode
Speaker 2

…has a positive effect on the multipliers before it. They stack. We believe that this can accelerate even further as we approach recursive self-improvement. Claude will be able to discover compute multipliers entirely by itself, allowing the number of compute multipliers to grow faster and faster.

Speaker 1

…week, but all you see is their emails. You can see how this would still be illuminating, but might not represent everything they have been reasoning about that week. Greg Brockman of OpenAI used to say he was excited by chain of thought monitorability. Now he says the situation doesn't look good.

Speaker 3

So the way that we've observed is the AI doing what we want it to do is through chain of thought monitoring. And when we first developed that technology, we said we have to make sure this stays faithful. But as the models have got more capable, they can do more kind of within just the forward path without having to actually kind of write out that thinking, which means we need a new approach to monitoring. We're going to need to deliver that. As the capabilities got to a certain point, chain of thought monitorability is degraded.

Speaker 1

Dare I say it, but for me, OpenAI are starting to seem more open in their communications. Far from a race to the top led by Anthropic, OpenAI's report yesterday says this. They admit that to maintain human oversight of systems, we will need new breakthroughs. Basically, on our current trajectory…

Speaker 4

I'm going to start with the most important. Sentence. So I think there is a high risk of an AI takeover within the next three years. What I mean by an AI takeover is a system situation where there are AI systems pursuing goals, and those systems call the shots in the same sense that humans today…

Speaker 1

Again, the point is you can call me skeptical that it's Anthropic leading this race to the top when it is OpenAI that chose to publish this crucial data. For what it's worth, they've also committed to providing even more information in areas such as the differential acceleration of capabilities…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from AI Explained Official Podcast