Discussions are underway among US Senate negotiators regarding a new artificial intelligence regulatory bill, according to reports from individuals familiar with the talks. The proposed legislation would require AI companies to provide proof that they have implemented reasonable safeguards to prevent their technologies from causing harm.

The push for stricter oversight gained momentum following the recent resignation of an Anthropic researcher, who cited concerns that developers of AI genuinely believe the technology could pose an existential threat before the end of this decade. This stark warning has intensified public scrutiny over the potential dangers associated with advanced AI systems.

In response to these mounting concerns, leaders from several major AI firms have called for a slowdown in technological development, a sentiment that contributed to a decline in global AI-related stocks the following day.

Key provisions under consideration

Negotiators are exploring giving the US Commerce Secretary expanded authority to demand that AI developers demonstrate compliance with their "duty of care" obligations, meaning they must show reasonable steps have been taken to prevent harm. Additionally, the Secretary could dispatch government auditors to directly test corporate AI products for safety and compliance.

However, the bill remains in the consultation phase and faces significant hurdles even if it were to pass through Congress, with expectations of enactment into law remaining low.

Judicial oversight and state law implications

Reports from last Friday indicate that negotiators are also considering involving federal courts in scenarios where the US government deems certain AI models unsafe and seeks to block their release. Under this framework, companies could challenge government decisions by filing objections with the federal judiciary. The extent of governmental authority and the specific procedural arrangements remain subjects of ongoing deliberation.

Furthermore, sources familiar with the matter indicate that the provisions related to federal courts may also prohibit states from enforcing their own laws concerning specific AI model risks, potentially centralizing regulatory oversight at the federal level.