Elon Musk says top US AI labs and “three or four of the leading Chinese companies” should let rivals run a “test harness” on their models to evaluate safety
Elon Musk called for the top artificial intelligence companies to work together and test each other's models …
CNBCAnnie Palmer
Context & Ripple Effects
Musk has argued for regulation of advanced AI since 2020, while the U.S. government established a narrower precedent in 2024 when OpenAI and Anthropic agreed to give the AI Safety Institute early access to major models for risk evaluations. His proposal shifts the emphasis from regulator access to reciprocal scrutiny among commercial rivals.
Musk’s proposal puts reciprocal safety access on the agenda for leading U.S. and Chinese AI labs; no commitment by any lab to provide that access is reported.
Labs would need test harnesses and audit-ready evaluation processes before rivals could assess their models under a shared framework.
Second-order effects
A voluntary peer-review model would force participating labs to balance safety validation against the competitive and security risks of exposing models to rivals.
The proposal offers an alternative to review centered on the U.S. AI Safety Institute, potentially making private labs more important in defining what constitutes an adequate safety test.
Third-order effects
If leading labs adopt reciprocal testing, access to frontier models could become a governance mechanism negotiated among companies and across national boundaries, rather than one controlled solely by regulators.
The durability of that model would depend on whether U.S. and Chinese participants can agree on disclosure limits and evaluation standards without turning safety review into a channel for competitive intelligence.
The trend: Frontier-AI governance is moving toward safety assurances that combine state oversight with industry-run evaluation, while model access becomes a geopolitical constraint.
Certainly a better solution than the 20 year prison sentence from the Bernie bunch... But what will be the motivation for the gatekeeping club to review the open source models? The analogy to peer review is not reassuring.
@chamath Agree. Grt to see these guys ideating in real time on the best path forward. Need to pragmatically balance safety concerns w speed, free markets & maximum competition. ⚖️🤖
Elon's idea of peer review is excellent and will incentivize each frontier lab company to invest in test harnesses and auditability. This is infinitely better than some transnational governance body to decide on humanity's future.
This is what they used to be called. It's what they should be called again. I polled my followers about what to call them and they liked the old name: supercompute it is!
So many great nuggets of information in this interview with Elon and Gwynne. 1. Elon rules out a slowdown on frontier AI development, instead proposing a cross-lab testing mechanism. I think it makes sense in theory. What's bullish though is that every fix Elon proposes uses more…
Elon Musk: Terafab is a Hedge Against Taiwan AND a Necessity to Scale Tesla/SpaceX @friedberg: “I'd love to hear the origin story of Terafab. What was the demand that made you say, we've got to do this, we've got to build this?” @elonmusk:
SITUATION EXPLAINED: Elon wants US and Chinese AI labs to test each other's models before release. • His proposal: instead of labs picking their own evaluators, they run safety test harnesses on each other • OpenAI tests SpaceXAI, SpaceXAI tests Anthropic, and so on, with Chinese…
It's a bit of a strawman to argue that a global treaty against ASI development would involve creating a ‘transnational governance body’. The International Atomic Energy Agency (IAEA) was established in the 1950s, and became the implementing body for the Treaty on the Non-Prolifer…
Grok Bot Summary of Elon Musk's All-In Summit Interview Today Opening bit - Joke cold open: “We're all going to die.” Death rate still 100%. - Then straight into “what happened in the last 72 hours.