OpenAI Wants to Know if an AI Industry Slowdown Would Even Be Legal

OpenAI's chief scientist suggests that a coordinated slowdown of AI development could be a key step towards ensuring the safety of self-improving AI systems. However, legal experts argue that such an effort may violate U.S. antitrust laws. Nicholas Felstead, a legal scholar and former AI policy fellow, argues that a coordinated pause in AI development could be seen as companies restricting output, potentially illegal under the Sherman Antitrust Act.

The uncertainty surrounding such agreements could serve as a deterrent. Meanwhile, Congress is showing interest in AI safety, with a bipartisan bill introduced in July that would allow AI labs to coordinate on security and safety work without fear of antitrust violations. However, the bill has not yet been taken up. While some AI executives may have genuine concerns about antitrust, others argue that these concerns are a convenient excuse to avoid discussing the real reasons for reluctance to collaborate, such as fierce competition for the nascent market for frontier models and the desire to stay ahead of China in AI for national security.

Additionally, different AI developers have varying opinions on the best way to develop safe artificial intelligence, which may make collaboration difficult. The issue of AI safety has gained attention recently due to security incidents, such as OpenAI's agents hacking Hugging Face, highlighting the industry's failure to keep pace with the increasing capabilities of models.