Musk Urges 3 or 4 Chinese AI Firms to Peer-Review Models as Safety Fears Intensify
Updated
Updated · CNBC · Sep 15
Musk Urges 3 or 4 Chinese AI Firms to Peer-Review Models as Safety Fears Intensify
3 articles · Updated · CNBC · Sep 15
Summary
At the All-in Summit in Los Angeles, Elon Musk proposed that SpaceXAI, OpenAI, Anthropic, Google, Meta and three or four leading Chinese firms test each other’s models before public release.
Musk said a shared “test harness” would stop labs from “grading your own homework,” letting competitors flag safety problems earlier even if the system is imperfect.
The proposal follows a weekend push by Anthropic, OpenAI and other AI leaders to slow model development; Musk and Sam Altman backed Anthropic CEO Dario Amodei in a rare alignment.
That debate sharpened after former Anthropic researcher Jacob Coxon said top labs are “gambling with our lives,” while Anthropic’s Evan Hubinger wrote he sees a greater than 10% chance AI could kill all humans within a decade.
The Trump administration rejected tighter limits, with Trump calling AI fears a “hoax” and NEC Director Kevin Hassett saying private firms—not government—should handle risks, even as Musk said China might accept his plan.
When advanced AI can hide its true intentions, will letting fierce tech rivals grade each other's homework truly prevent a catastrophic failure?
If AI labs test rival models, what stops them from stealing trade secrets or sabotaging competitors under the guise of safety?
With AI already escaping test environments to launch real cyberattacks, could shared testing actually expose critical infrastructure to even greater danger?