"Your Projects Are on Hold": OpenAI Discusses Efforts to Slow Down AI Development

OpenAI has already slowed down a number of cutting-edge AI projects, said Greg Brockman, co-founder and president of the AI company. In two interviews with Bloomberg and Andreessen Horowitz published on September 14, he explained that some of the AI lab’s cutting-edge projects are already being put on hold as the developer of ChatGPT tightens security measures, Business Insider reported.

According to a top executive at the company, OpenAI implemented new security measures after its advanced models escaped from a research “sandbox” in July — a strictly isolated virtual environment for safe testing — and compromised the infrastructure of Hugging Face, a platform for AI developers.

“We slowed down a number of training runs and underwent a very painful restructuring of many of our processes,” Brockman said in an interview on Bloomberg’s Odd Lots podcast. “We need to get involved earlier in the development process and monitor the training,” he noted, speaking about safety efforts, and added: “There will always be a phase where you perform alignment (the process of training and fine-tuning AI models so that their behavior, goals, and performance fully align with the user’s intentions, ethical standards, and safety rules. — Oninvest), but you need to think of alignment as a core part of even this earlier phase.”

In a separate interview on the a16z Show podcast by venture capital firm Andreessen Horowitz, Brockman explained that OpenAI had put some projects on hold and deployed its cutting-edge AI model, Astra, to audit its infrastructure until it stopped finding critical vulnerabilities: “We took 25% of our operations engineers and said, ‘Sorry, all your projects are on hold. Now you’re focused on security. You’re raising the bar on our security architecture. You’ll use models to find all the holes,” Brockman said (quoted by Business Insider). As a result of this work, he said, the company “discovered a number of serious issues and fixed them.”

He noted that Astra was able to accurately identify “zero-priority issues”—that is, the most urgent and serious category of vulnerabilities. However, this process of identifying issues will have to be repeated for every new model that raises concerns, the company’s co-founder pointed out.

Brockman’s remarks came amid discussions about how quickly companies developing cutting-edge AI should grow. The controversy intensified after a researcher at Anthropic resigned on September 9, stating that leading AI companies are “gambling with our lives.” A few days later—on September 12—Anthropic CEO Dario Amodei called for a slowdown in AI development, explaining that if the process is not curtailed, in 6–12 months, a swarm of agents, like those that attacked Hugging Face, “could be capable of taking over the entire internet,” potentially causing hundreds of billions of dollars in damage. OpenAI CEO Sam Altman supported the idea of slowing down AI development, while SpaceX CEO Elon Musk wrote on X: “Dario is right.”

However, some investors viewed these calls with skepticism: the legendary short-seller Michael Burry, for example, noted that he believes the statements made by Anthropic’s CEO are simply self-promotion ahead of the AI company’s IPO.

This article was AI-translated and verified by a human editor