Elon Musk is calling for the world’s biggest artificial intelligence companies to test each other’s most powerful models before they reach the public.
Speaking at the All-In Summit in Los Angeles on Monday, Musk proposed that xAI, OpenAI, Anthropic, Google, Meta and several leading Chinese AI companies allow rivals to run a “test harness” against their models before release.
Rather than companies effectively “grading their own homework,” Musk argued that competitors could search for dangerous capabilities or weaknesses and raise alarms when problems emerge. He acknowledged that the system would not eliminate every risk, but said it could dramatically increase the chances of discovering them.
AI Safety Debate Intensifies
The proposal comes during a sharp escalation in warnings from the AI industry itself. Anthropic CEO Dario Amodei called for more caution around increasingly capable frontier models, while Musk and OpenAI CEO Sam Altman have also backed calls for stronger safeguards.
However, the industry is still very divided over whether slowing development is the right response. Meta CEO Mark Zuckerberg argued that AI laboratories already have strong incentives to build safely, while Nvidia CEO Jensen Huang also pushed back against efforts that could slow technological progress.
Musk’s idea would not start from scratch. The Frontier Model Forum, whose members include OpenAI, Anthropic, Google and Microsoft, has already developed guidance for third-party assessments of advanced models. The US National Institute of Standards and Technology has also expanded efforts to independently evaluate AI systems.
Musk’s proposal goes a step further by putting competing frontier labs directly inside the testing process.
The debate is also exposing a growing divide over who should regulate advanced AI. President Donald Trump rejected calls for a broad slowdown, arguing that restrictions on US AI development could hand China an advantage.
At the same time, Musk’s own xAI has challenged state-level AI regulations, including laws governing AI-generated images and training-data transparency.
That creates an important contradiction at the heart of the debate. Musk appears to favor stronger scrutiny of powerful AI models, but through industry-led mechanisms rather than government restrictions.
