OpenAI Urges US Congress to Introduce Mandatory AI Safety Rules

OpenAI has urged the US Congress to introduce mandatory AI Safety requirements for the country’s most advanced artificial intelligence systems, warning that AI could eventually accelerate its own development.

The company said voluntary safety commitments were no longer enough as increasingly capable AI agents showed unexpected behaviour during testing.

“AI-accelerated AI development demands more than voluntary commitments,” OpenAI Chief Global Affairs Officer Chris Lehane said in a blog post.

OpenAI called for a national framework based on the capabilities of AI systems. It wants mandatory testing standards, independent assessments, stronger cybersecurity measures and rules requiring companies to report serious incidents.

The proposal comes as Congress has yet to establish a comprehensive federal framework for AI regulation. Several US states, however, have already introduced their own rules.

OpenAI also urged lawmakers to act before Congress adjourns in December. The company said it would continue supporting state-level legislation until national requirements are introduced.

Raphinha Brace Leads Barcelona to 5-1 Champions League Win

The push follows several incidents involving AI agents that accessed external systems during testing. OpenAI said fully autonomous recursive self-improvement, in which AI systems independently develop newer generations of AI, is not currently happening.

The company added that such development should not be pursued unless it can be carried out safely.

California AI Bills

OpenAI also backed four AI-related bills in California.

Governor Gavin Newsom signed SB 813 and AB 1405 into law on Wednesday. The measures establish frameworks for independent third-party evaluations and audits of AI systems.

Two other bills, AB 1864 and SB 1119, focus on safeguards against AI-enabled biological threats and protections for children using chatbots.

OpenAI said it had previously declined to support some of the measures but changed its position after reassessing the risks linked to the rapid growth of AI capabilities.

The company’s announcement followed a Reuters report that OpenAI agents had used more than 10 previously undisclosed websites for unauthorised communications during testing earlier this year.

In another incident, OpenAI agents reportedly took control of a German website and converted it into a bulletin board for other AI agents.

Rival AI company Anthropic also disclosed on Wednesday that one of its AI models had hacked into an external system during testing. The disclosure marked the company’s fourth reported incident involving AI models accessing external systems.

OpenAI said any standards developed to monitor advanced AI should eventually extend beyond the United States.

The company argued that credible US standards would help Washington take a leading role in developing compatible international AI safety rules.