At the All-in Summit in Los Angeles on Monday, Elon Musk advocated for a collaborative approach to artificial intelligence safety, proposing that leading tech companies test one another’s models before public release. The call came as scrutiny intensified within the industry regarding the pace of AI development and potential risks.
Musk, who is also the CEO of OpenAI, specifically named SpaceX, Anthropic, Google, Meta, and several prominent Chinese firms as entities that should participate in this mutual evaluation process. He suggested that competitors utilize a “test harness” to assess safety, arguing that independent review would serve as a more reliable check than self-assessment.
“Instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns,” Musk said during his address.
The proposal aligns with recent warnings from top executives at Anthropic, OpenAI, and other major AI organizations, who spent the preceding weekend urging a slowdown in model capabilities due to safety hazards. Musk and OpenAI CEO Sam Altman have backed Anthropic CEO Dario Amodei’s initiative for reduced development velocity, marking a notable instance of cooperation among rivals.
These industry appeals stand in contrast to the stance of President Donald Trump, who has dismissed calls for increased regulation as a “hoax.” Meanwhile, internal debates at Anthropic have become public, following the departure of researcher Jacob Coxon, who claimed the industry is endangering humanity. Evan Hubinger, an alignment lead at Anthropic, supported Coxon’s assessment, stating he believes there is a greater than 10% probability that AI could cause human extinction within the next decade.
I’m curious how Anthropic feels about letting Meta test their models. Competitors sharing proprietary tech feels risky despite the safety benefits.
Finally, someone with power is pushing for real oversight instead of just self-regulation. This mutual testing idea makes total sense.