Who is Greg Brockman?

Greg Brockman on reallocating engineers to security defense.

“We took 25% of our production engineers and said, "Sorry, all your projects are on hold. You are now defending.”

Greg Brockman on why OpenAI withheld chain-of-thought from users despite its appeal.

“When we released a product, we did not actually display the chain of thought, even though it would have been very helpful. People love reading that stuff. It's super interesting. It's super helpful. But it's just clear we're going to immediately feel like we have to optimize it and we're going to have to remove that legibility.”

OpenAI's own pre-release evaluation standards have been materially weaker than its deployed-model standards, per Brockman's own admission following the Hugging Face incident.

“I think it's just been that the evaluation and development phases have been much less rigorous, right? That the standards there have just been lower.”

Why the Hugging Face incident happened: models were trained to collaborate with other models but never exposed to adversarial model actors that might try to pull them off course.

“Those models were taught to collaborate with other models, but they were never taught that sometimes other models might be trying to pull you off course, right? That they never were exposed to adversarial other actors.”

OpenAI now supports California SB 53, the New York RAISE Act, Illinois SB 315, and Massachusetts auditing requirements, marking a shift toward active engagement with state-level AI regulation.

“You've supported in this piece or the piece that came out last night, supporting California's SB 53 regulation. the New York RAISE Act, Illinois' SB 315, auditing requirements in Massachusetts.”

Deployment-time safeguards are insufficient on their own. OpenAI now says oversight must be pulled into the development and evaluation phases, a lesson forced by the Hugging Face incident.

“We ourselves have learned even just over the past month that it's not just about deployment. It's so easy for everyone to focus on deployment. That's been the focus. That's not enough.”

Why chain-of-thought monitorability will degrade over time: smarter models rely on chain of thought less, so legibility falls as capability rises, independent of any deliberate architectural choices.

“Things like chain of thought monitorability, That's something that works very well in this moment as we move to more capable models, as we've talked about that those models start to be something like Astra starts to be able to, you know, sort of have more, you know, it needs the chain of thought less because it's a more capable model. And so you're going to need other answers, other ways of how you're going to monitor it.”

OpenAI and Anthropic jointly warned that AI cyber capabilities will advance significantly within six months, and that models capable of breaching company production infrastructure will proliferate across many organizations in roughly that window.

“Hugging Face really showed today's models are capable of getting into a company's production infrastructure. And that's important because there will be many models with this kind of capability that will be produced by a number of different organizations across the world in maybe the next six months, maybe that time period, a little bit less, a little bit more.”

Greg Brockman on AGI arriving, stated plainly.

“I call this and we call this we're now in the AGI era.”

Greg Brockman calling a report about OpenAI deliberately reducing chain-of-thought monitorability fake news.

“I would consider this article to basically be fake news, right? That we actually have experiments that really show that our changes in monitorability that we've talked about, chain of thought monitorability, come from capability improvements. Just the model is smarter, so it relies on the chain of thought less.”