Elon Musk: test labs’ models against each other as a practical alternative to a universal pause
CNBC ● Covered by 27 sources
Musk wants AI rivals to test each other’s models before launch. He says peer review beats everyone grading their own homework, but the labs still haven’t signed on.
Based on reporting by CNBC — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Elon Musk is pitching a pretty simple fix for one of AI’s messiest problems: let the big labs check each other’s work before the models go public. Speaking at the All-In Summit in Los Angeles on Monday, he said xAI, OpenAI, Anthropic, Google, Meta and several leading Chinese companies should all allow rivals to run a “test harness” on their systems to spot safety issues early.
His logic was blunt. If competitors are the ones grading the homework, he said, they’re more likely to notice problems and sound the alarm. Musk added that the idea is not perfect, but argued that it would make it much more likely that serious flaws get caught.
The call lands in the middle of a louder-than-usual fight over AI risk. Over the weekend, leaders from Anthropic, OpenAI and other companies warned about the harms of the technology and backed a slowdown in model development. Musk and OpenAI chief Sam Altman were among the people who supported Anthropic CEO Dario Amodei’s slowdown proposal, which is not something the AI world does every day.
That rare unity came after another alarm bell: a former Anthropic and OpenAI employee, Jacob Coxon, said the leading AI labs are “gambling with our lives.” Anthropic’s Evan Hubinger went even further, saying he personally thinks AI could kill all humans and put the odds above 10% within the next decade.
The politics around that warning are already ugly. The Trump administration has brushed off calls for tighter limits, with Donald Trump calling AI fears a “hoax” and a “scam,” while Kevin Hassett said the private sector is the right place to handle the issue. Musk’s own companies have also fought state AI rules in court, even as xAI faces probes and lawsuits tied to Grok image generators used to create deepfake porn and other nonconsensual images. So yes, the man is calling for self-policing — while his own firms are busy suing the referees.
My take — AI-written commentary, not fact-checked reporting
This is the AI industry’s favorite trick: ask for “responsible” oversight, then fight the actual rules the moment they show up. Peer review sounds fine, and it’s better than everyone circling the wagons, but it only works if the labs stop treating safety like a branding exercise. Open models, closed models, whatever — the real divide is between people who want scrutiny and people who want permission slips.
Read more about this at: CNBC