OpenAI urges mandatory US AI safety rules

OpenAI asked Congress to create binding, capability-based safety requirements for the most advanced AI systems. The company said voluntary pledges no longer suffice as AI becomes more autonomous. It proposed shared thresholds for slowing or stopping development and said rules should focus on well-resourced frontier laboratories.

• 6–12 monthsAnthropic’s cited risk horizon — Estimate from its chief executive
• Several daysTasks some agents can complete — Compared with skilled researchers

Q1

What would the rules require?

OpenAI’s proposal includes common testing, independent assessments, stronger cybersecurity protections and mandatory reporting of serious incidents.

Q2

Why target only frontier laboratories?

OpenAI said limiting rules to well-resourced developers would focus safeguards on systems with capabilities it considers highest risk.

Q3

How advanced are current AI agents?

OpenAI said some agents already complete tasks that would take skilled researchers several days, accelerating research and alignment work.

Q4

What marks a change in policy?

OpenAI acknowledged that some safeguards it now supports were measures the company had previously declined to endorse.

• ABC News
• Indiatimes
• Anadolu Agency
• RNZ
• Handelsblatt
• CNN
• The Guardian
• CNBC