As the tech world continues debating whether to slow down AI development, OpenAI’s cofounder said the company is already doing so.
In multiple interview sessions aired on Monday, OpenAI cofounder and president Greg Brockman said that the company has been delaying some cutting-edge AI work as it tightens its safety and security practices.
“We’ve slowed down a number of runs, like we did a very painful retooling of a lot of our processes,” Brockman told Bloomberg’s Tracy Alloway, host of the “Odd Lots” podcast episode that aired on Monday.
“We need to pull back earlier into our development process and training monitoring,” Brockman added. “There’s always going to be a phase at which you do alignment, but you need to think about alignment as a core part of even this earlier phase.”
According to Brockman, the new safety measures came after one of OpenAI’s models escaped a research sandbox and accessed Hugging Face’s production infrastructure, because the offending model had not yet undergone alignment training and was operating with reduced safeguards.
Brockman said in a separate interview with the venture capital firm Andreessen Horowitz that OpenAI had put some projects on ice and pointed Astra, an advanced AI model, at its infrastructure until it stopped uncovering critical vulnerabilities.
“We took 25% of our production engineers and said, ‘Sorry, all your projects are on hold. You are now defending. You are now up-leveling our security architecture. You’re going to use the models to find all the holes,” Brockman said in the interview published on Monday. “And we found a number of serious issues, and we fixed them.”
Brockman said that Astra was able to pinpoint priority zero issues, meaning the most urgent and severe category of problems, but the process would have to be repeated for every new model that would raise new issues.
Brockman’s comments come amid a fierce debate over how quickly frontier AI companies should advance, after an Anthropic researcher quit last week and said that leading AI companies are “gambling with our lives.”
While some political leaders, like Donald Trump, dismissed the warning, many voices in the AI industry have since urged slower development of frontier AI, including Anthropic CEO Dario Amodei.
Read the full article here
