Elon Musk proposes mutual model testing for AI labs

At the All-In Summit in Los Angeles on September 14, Elon Musk proposed that xAI, OpenAI, Anthropic, Google, Meta, and leading Chinese companies test competitors' models on a shared test harness before public release, stating that the chances of finding problems would become significantly greater.

Compiled by Sergey KostenchukPublished 2026-09-15Updated 2026-09-18

2026-09-15 Safety OpenAI

🛡 Elon Musk proposes mutual model testing for AI labs

At the All-In Summit in Los Angeles on September 14, he called on xAI, OpenAI, Anthropic, Google, Meta, and three or four leading Chinese companies to give rivals a shared test harness before release — a platform where competitors themselves look for vulnerabilities in each other's models. The method is not perfect, but "the chances of finding problems will become significantly greater," Musk said.

🌍 Mutual testing turns competition into a safety mechanism: rivals will find what the developers themselves miss — this is an alternative to government regulation. But no lab has agreed yet, and the race with China hinders voluntary slowing down.

👤 The debate "AI is dangerous, we need to slow down" is moving into practical proposals: Dario Amodei's essay on a pause was supported by Sam Altman and Musk himself. Meanwhile, Trump calls AI concerns a "hoax," and China's Foreign Ministry calls it "fear mongering."

Source 1: https://www.cnbc.com/2026/09/15/elon-musk-ai-safety-testing.html

Sources