At the All-In Summit in Los Angeles on Monday, Elon Musk made a proposal that is unusually concrete for this debate: xAI, OpenAI, Anthropic, Google, Meta and “three or four of the leading Chinese companies” should let each other at their models before those models ship.
His word for it is a “test harness” — a rig a competitor hooks up to somebody else’s model.
The line that carries it
“So, you know, instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns.” Not a perfect solution, Musk conceded. “The odds that you will find issues are dramatically greater.”
Nobody has agreed so far. Musk says so himself. He still thinks it is doable, and quickly: “I think it’s probably something that China would agree to.”
Where this fits in the week
It is the third version of the same idea in seven days. Dario Amodei proposed embedded evaluators — independent third parties sitting inside the building, verifying safety practice. Sam Altman explicitly welcomed independent evaluators yesterday. Musk now suggests skipping the evaluators and letting the competition in instead.
Washington answered on Tuesday, more or less as expected. Kevin Hassett, director of the National Economic Council, told CNBC the private sector is “the right place” to deal with these concerns; the government keeps watching and will use “law enforcement when necessary”. Beijing had called the AI bosses’ warnings fear mongering on Monday.
The cheapest proposal on the table
Of everything floated this week, Musk’s version has the fewest prerequisites. Embedded evaluators need an institution, staff and some kind of mandate. A law needs Congress, and Congress leaves for the campaign trail this week. Mutual testing needs two companies that agree to it.
Which is exactly the catch. Show a competitor your model before release and you hand them a view of its capabilities, its weaknesses and your timeline. Musk folded xAI into SpaceX in February and has been fighting with OpenAI for years. That he is the one making the proposal is the weakest possible argument for anyone accepting it.
Which leaves the question he raises himself. If the proposal is this cheap and still nobody takes it — what is that telling us?