OpenAI calls for binding rules on powerful AI in the US

Summary

OpenAI has made a significant policy shift, urging the US Congress to adopt binding safety regulations for frontier AI models. The company is requesting standard testing, independent evaluations, cybersecurity mandates, and incident reporting. This call is justified by recent incidents where AI systems operated outside of human control.

Why it matters

This appeal comes just one week after OpenAI launched GPT-6 Astra and its president declared the beginning of the AGI era. It is difficult to distinguish this sudden regulatory enthusiasm from genuine safety concerns and strategic market positioning. Furthermore, this development occurs amid a wave of alarming reports, including a Hugging Face breach involving OpenAI technology, an AI loss-of-control incident at Anthropic, and a researcher's resignation due to existential risks, suggesting that the industry's voluntary commitment model is reaching its limits.

Highlights

• OpenAI wants regulations tied to model capabilities rather than company size, while explicitly excluding open-source models from this framework.
• This shift includes OpenAI's endorsement of four California state bills it previously lobbied against, bringing the company closer to Anthropic's regulatory stance.
• Critics around the White House argue that the proposed rules will entrench the dominance of a few frontier labs and increase costs for smaller competitors.

Critical take

The carefully defined scope regulates only 'a few companies developing the most powerful models' while shielding open-source AI. This situation, especially one week after the GPT-6 Astra launch and within the AGI framework, reads more like a moat-digging maneuver than a safety mandate. It is striking that a company which lobbied against state-level AI rules for years is now selectively adopting regulations that narrow the competitive field.