Top AI Leaders Have Warned About The Tech Moving Too Fast. They Have Now Started Working Together.
OpenAI has begun working with other top AI companies to address the potential dangers that the technology could pose if not paced.
The company’s policy chief, Chris Lehane, told Bloomberg News that the company is working with Anthropic and Alphabet on AI Safety.
He went on to say that the companies have been in talks for several weeks and doesn’t believe they need an antitrust waiver to coordinate. “It’s better to try to work together to prioritize safety,” he said.
The remarks follow a report claiming that the companies were also discussing the creation of a new industry-wide AI safety body that could establish common standards for the world’s most powerful artificial intelligence systems.
The private discussions among the three companies have been underway as their leaders publicly call for slowing the development of increasingly capable AI, The Information detailed.
The potential organization could become a central mechanism for setting safety standards across companies developing frontier AI. It could also provide a framework that allows competitors to coordinate on safety without forcing individual companies to slow down while rivals continue racing ahead.
Such a body could potentially establish checkpoints that AI developers would have to meet as their models become more powerful. Amodei has proposed tying increasingly advanced capabilities to corresponding safety requirements, including evaluations, interpretability testing, and audits of the environments used to train models.
The move comes just days after Anthropic CEO Dario Amodei called for slowing the pace of AI advancement, a position backed by OpenAI CEO Sam Altman and Elon Musk.
Another major component would be independent oversight. Amodei announced that Anthropic would give third-party safety evaluators ongoing, “employee-like access” to parts of its operations. Those evaluators would be able to verify whether the company is following its safety commitments, assess training processes and report serious incidents. OpenAI CEO Sam Altman subsequently said his company would make a similar commitment to independent evaluators.
Musk made a similar call for developing safety measures on Monday, saying at the All-in Summit in Los Angeles that the companies, including “three or four of the leading Chinese companies” should let competitors run a “test harness” on their models to evaluate them.
“So, you know, instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns,” Musk said.
