Elon Musk proposed that the world's major artificial intelligence companies should first undergo peer testing before making their strongest models available to the public. This statement came from his speech at All - In Summit in Los Angeles, and it involves xAI, OpenAI, Anthropic, Google, Meta, as well as some Chinese AI companies.
He suggested that competitors run a set of unified testing tools to check the models before release, rather than having the developers assess them on their own. Musk stated that this approach cannot eliminate all risks, but it can significantly increase the likelihood of detecting dangers and system weaknesses.
AI Security divergence continues to widen
At the time this proposal emerged, discussions within the AI industry about the risks of cutting-edge models were intensifying. Dario Amodei, the CEO of Anthropic, recently advocated for greater vigilance regarding continuously improving cutting-edge models. Elon Musk and OpenAI, the CEO of Sam Altman, also support the strengthening of security measures.
However, the industry has not formed a unified stance. Mark Zuckerberg, the CEO of Facebook, believes that the laboratories themselves have sufficient motivation to ensure product safety. Jensen Huang, the CEO of NVIDIA, also opposes practices that could slow down technological progress.
The existing evaluation mechanism is already in progress.
Elon Musk's proposals are not entirely starting from scratch. The Frontier Model Forum initiative, which involves institutions such as OpenAI, Anthropic, Google, and Microsoft, has already proposed third-party evaluation guidelines for advanced models. The National Institute of Standards and Technology (NIST) in the United States is also expanding its independent evaluation work on AI systems.
However, Musk's approach goes a step further by emphasizing the involvement of directly competing cutting-edge laboratories in the testing phase, allowing peers to identify issues and issue warnings.
Differences between industry self-regulation and government supervision become apparent
Within the industry, there is also a division over who should be held to stricter standards. U.S. President Donald Trump opposes widespread slowdowns in AI development, believing that restricting the development of the U.S. AI could give China an advantage.
At the same time, xAI under Musk has also opposed state-level AI regulatory measures, including laws related to the transparency of AI image generation and training data. This outlines a clear stance for Musk: he supports stricter scrutiny of powerful models, but prefers industry-led approaches rather than relying on government restrictions.
