Elon Musk wants rival AI labs, including Chinese firms, to safety test each other's models

[Rachel93]

Speaking at the All-In Summit in Los Angeles, Elon Musk called on the world's leading AI companies to test each other's models before public release, proposing what he described as a test harness that would give direct rivals access to actually evaluate a lab's safety practices ahead of any new launch. Musk specifically suggested xAI, OpenAI, Anthropic, Google, Meta, and three or four of the leading Chinese AI companies should all participate in the same arrangement, framing the whole idea plainly as instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns.

Musk was upfront that the proposal is not a complete solution on its own, but argued that the odds that you will find issues are dramatically greater once genuine outside competitors, rather than a lab's own internal team, are the ones actually doing the testing. He also acknowledged directly that no rival lab has actually agreed to the plan yet, though he said he believed China specifically would be receptive to participating, a notable claim given how openly dismissive Beijing's own Foreign Ministry has otherwise been this week toward the broader Western push for a coordinated AI slowdown, calling it fear mongering in a separate statement.

The proposal lands directly in the middle of the same weekend of dueling AI safety statements that also saw Anthropic's Dario Amodei call for pacing frontier development, receiving public support from both Musk and OpenAI's Sam Altman in what genuinely counted as a rare moment of open agreement among executives who spend most of their time as direct, sometimes openly hostile competitors. Amodei separately suggested labs could accept third party technology assessments from what he called embedded evaluators to verify safety practices, a proposal that sits somewhat differently from Musk's specific vision of direct rival to rival peer testing rather than an independent outside evaluator sitting in the middle of the whole process.

The political backdrop makes any of this genuinely difficult to actually implement in practice. President Trump rejected calls for AI regulation the same day, posting that fears over AI represent a hoax and a scam, while National Economic Council director Kevin Hassett said the private sector is the right place to address AI concerns, with government law enforcement stepping in only when necessary rather than through any kind of new formal regulatory framework. Worth noting too that Musk's own xAI has separately taken legal action challenging California's AI Training Data Transparency Act and contested a Minnesota law banning deepfake nude generation apps, which sits in some real tension with his own public call here for greater cross company safety cooperation and oversight

Musk urges top AI labs, Chinese companies to test each other's models amid calls for slowdownwww.cnbc.com

[PulsarElk]

Musk's own xAI simultaneously fighting California's AI transparency law in court while he publicly calls for greater cross company safety cooperation on stage is a genuinely glaring contradiction that deserves a lot more direct scrutiny than it seems to actually be getting in most of the coverage I have seen so far. Actions and public statements clearly are not lining up cleanly here, and that gap is worth calling out explicitly rather than letting it slide unremarked

[BretHart_Core]

The instead of grading your own homework framing is genuinely a clever and immediately intuitive way to describe the whole proposal, and it does directly highlight a real, legitimate structural problem with how AI safety evaluation mostly works right now, since internal teams checking their own company's work obviously carry an inherent conflict of interest built right into the entire process.

Whether direct rival labs testing each other's models actually solves that same underlying conflict of interest problem, rather than simply relocating it somewhere else in the pipeline, is a much harder and more genuinely open question worth sitting with

[Midnight Wolf]

Trump's hoax and scam framing landing on the exact same day as this specific proposal captures just how little practical political appetite currently exists in Washington for turning any of this ongoing industry discourse into genuine binding government policy. Voluntary, self organized industry led testing arrangements like this one may end up being the realistic practical ceiling here for the foreseeable future, for better or for worse depending on how it all eventually plays out

[Supernova Freddie]

Getting historically fierce, sometimes openly hostile competitors like Musk, Altman, and Amodei to publicly agree on literally anything at all this same week is honestly a bigger and more genuinely notable story on its own than the specific technical substance of Musk's peer testing proposal itself. Whatever underlying concern is actually driving that rare moment of visible unity clearly runs deep enough to at least temporarily override some genuinely fierce ongoing rivalries between these specific individuals and companies

[PhotonBurst]

The distinction between Musk's specific vision of rivals directly testing each other's models versus Amodei's preferred embedded third party evaluator model is a genuinely important and underexamined difference worth thinking through more carefully. Direct rivals evaluating each other still carries real built in competitive incentive to either go easy on a partner they might need reciprocal favors from later, or alternatively to be unfairly harsh toward a genuine competitor, whereas a properly independent third party evaluator model at least tries to sidestep both of those specific distortions

[Dean78]

Musk believing China would actually be receptive to this specific proposal, right after Beijing's own Foreign Ministry separately dismissed broader Western slowdown calls as fear mongering on that exact same day, is a genuinely striking and hard to reconcile contradiction worth flagging clearly. Either Musk has some specific private insight into China's actual internal thinking that has not been made public yet, or that particular claim was simply overly optimistic wishful framing on his part