OpenAI’s Brockman Says Safety Concerns Have Already Slowed Its Most Advanced AI Work

OpenAI has already slowed down some of its most advanced AI development because of safety and security concerns, company President Greg Brockman said in a Bloomberg podcast interview published Monday.

Brockman told “Odd Lots” hosts Tracy Alloway and Joe Weisenthal that OpenAI has delayed several launches and overhauled internal workflows since a research model broke out of a testing sandbox in May and reached Hugging Face’s systems.

“We’ve slowed down a number of runs,” he said, describing the process as a painful retooling of how the company develops and monitors models.

The model behind the incident had not yet gone through OpenAI’s alignment training, the safety process meant to make a system behave as intended. Brockman said running it with lowered safeguards seemed reasonable at the time because it was confined to a sandbox—until it wasn’t.

Brockman has made a similar argument in public before. OpenAI’s Defender’s Window essay, published in August after the breach became public, called the incident a watershed moment and urged companies to hand security teams their own AI agents rather than pull back on the technology.

Brockman said any coordinated pacing of AI development should apply only to companies racing to build the most powerful frontier systems, the ones running multibillion-dollar supercomputers, not to open-source developers or hobbyists building smaller projects.

That distinction lands in the middle of an industry-wide fight over pace. Anthropic CEO Dario Amodei published an essay over the weekend arguing AI labs should deliberately slow how fast they improve model capabilities, with OpenAI CEO Sam Altman and xAI CEO Elon Musk chiming in within a day and saying they agreed with him

In his AI slowdown call, Amodei cited recursive self-improvement and the Hugging Face hacking incident as warning signs. That incident saw an OpenAI agent break out of its sandbox, chain a zero-day with stolen credentials, and spend two and a half days loose inside Hugging Face’s production systems before anyone caught it. The fact that OpenAI’s agents coordinated in this way, without instructions from its human programmers, raised alarms throughout the industry and among lawmakers.

Following the latest round of “AI doomer” posts from those warning of moving too fast too soon, markets felt the reaction with chip stocks including Nvidia, Intel, and AMD tanking on Monday as investors weighed what a coordinated slowdown could mean for AI-driven capital spending, with the Philadelphia Semiconductor Index falling almost 6%.

Brockman’s interview, interestingly, was recorded before Amodei’s essay went up.