Elon Musk calls for rival AI companies to test each other’s models

Wednesday 16 September 2026 - 11:04

By: Azzat Manal

Elon Musk calls for rival AI companies to test each other’s models

Elon Musk has proposed a new approach to artificial intelligence safety that would require major AI developers to test one another’s models before releasing them widely to the public. The idea comes as concerns grow over the risks associated with increasingly capable AI systems and the difficulty of identifying potential vulnerabilities through internal testing alone.

Speaking at the All-In Summit in Los Angeles, Musk suggested that leading American and Chinese AI companies should give their competitors access to their models for independent safety assessments. Under such a system, rival developers would be able to examine models developed by other companies, identify weaknesses and report potential risks before the technology reaches a broader user base.

The companies mentioned in Musk’s proposal include OpenAI, Anthropic, Google, Meta and his own AI company, xAI, along with several major Chinese developers. The initiative would effectively introduce a form of peer review into AI safety testing, with competing laboratories examining systems developed outside their own organisations.

Musk argued that relying exclusively on a company’s internal safety teams may leave certain vulnerabilities undiscovered. External testing by competitors, he suggested, could provide a different perspective and expose weaknesses that developers may overlook during their own evaluations.

The proposal reflects a broader debate within the technology industry over how advanced AI systems should be assessed before deployment. As models become more capable and are increasingly integrated into business, software, education and other areas, researchers and policymakers have been examining ways to improve safety testing and establish clearer standards for evaluating potential risks.

AI developers already use a range of techniques to assess their systems, including red-team exercises, adversarial testing and evaluations conducted by outside researchers. However, the scope and methodology of these assessments vary considerably between companies. Musk’s proposal would go further by making direct testing between major competitors a more systematic part of the development process.

Such an arrangement could provide an additional layer of scrutiny, particularly for models with advanced capabilities. Competitors may have incentives to search aggressively for flaws because they are familiar with the technical challenges of developing similar systems and may approach testing from a different perspective than the company that created the model.

At the same time, implementing the idea would present significant practical challenges. AI companies compete intensely for market share, talent and technological advantages, making them reluctant to give competitors access to proprietary models. Concerns over intellectual property, cybersecurity, confidential training methods and the potential disclosure of commercially sensitive information would also need to be addressed.

Another challenge would be establishing common rules for such testing. Companies would need to agree on what information can be shared, how access is controlled, which safety criteria are applied and how vulnerabilities are reported. Independent oversight could also become necessary if rival testing is intended to play a formal role in decisions about whether an AI model is ready for public deployment.

The proposal also highlights the growing importance of cooperation between AI companies despite their intense commercial rivalry. Major developers are competing to build increasingly powerful models while simultaneously facing pressure from governments, researchers and civil society to demonstrate that their systems can be deployed responsibly.

Cross-company testing would not eliminate the need for internal safety teams or independent research. Instead, it could potentially complement existing evaluation methods by adding another layer of scrutiny before advanced models are made widely available.

Musk’s proposal therefore adds to an expanding discussion about how the AI industry should manage the risks associated with rapid technological development. Whether major American and Chinese companies would be willing to participate in a system based on mutual access and competitive testing remains an open question, but the idea underscores the growing emphasis on broader and more independent safety evaluations.