WASHINGTON, D.C. – In case you missed it, in a new op-ed published in The Washington Post, Rep. George Whitesides (D-CA), a member of the House Science, Space and Technology Committee and former chief of staff of NASA, called on the leading AI developers to pause work on their most advanced models for 30 days and adopt shared safety protocols that then would be backed by Congressional legislation.

Drawing on his background in aerospace, where a serious flight accident triggers an industry-wide “safety stand-down,” Rep. Whitesides argued that AI companies should apply the same discipline to frontier model development. The op-ed comes days after the leaders of Anthropic, OpenAI and xAI came out in favor of slowing frontier AI development, and follows reports of an OpenAI model breaking out of its digital enclosure.

Read Rep. Whitesides’ key points below:

The Washington Post (Opinion): How to keep AI from going rogue? Start with a 30-day stand-down.

By Rep. George Whitesides

• I spent my career in aerospace before coming to Congress, dealing with high-stakes missions with life or death outcomes. In my old industry, a serious flight accident triggers a safety stand-down: You temporarily stop flying or launching, study what happened, document what you will do to fly safely going forward and then return to flight. 
  
• The frontier developers – OpenAI, Anthropic, Google, xAI, Meta and potentially a few others – should take a lesson from the aerospace industry. They should conduct a safety stand down together and pause work on their most advanced models for 30 days. This would not stop anyone’s access to current models. What stops is training the next frontier model and, above all, using AI systems to accelerate AI research on themselves. 
  
• During those 30 days, experts from the companies should work with independent experts appointed by the government to accomplish one goal: agree on red lines for existential risks and recursive self-improvement, and mechanisms to monitor and verify.
  
• Stopping work on superintelligence is not likely to be mandated anytime soon by a Congress that has yet to pass a frontier-AI law, and an unconstrained pause invites the race-with-China concern. A voluntary one-month safety stand-down, called for publicly by Congress so as to put antitrust lawyers at ease, asks the companies only to do together what OpenAI did alone in August, when it halted part of its training for two weeks.

• Congress should use the same month to draft the thresholds and verification structure – independent auditors, government observers inside the labs – that turn voluntary red lines into law. 

• The people building this technology put the odds of catastrophe somewhere between 1 in 100 and a coin flip. You would not board a plane that had those odds of survival. 

You can read the full piece in the Washington Post here.

###