Should AI Companies Test Each Other’s Models?

Artificial intelligence is advancing at an extraordinary pace, and with increasingly powerful AI models comes a growing question: Who should test the technology before it reaches the public?

At the All-In Summit in Los Angeles this week, Elon Musk proposed an unusual approach: let AI companies test each other’s models before they are released. His proposal would involve leading AI companies including xAI, OpenAI, Anthropic, Google, Meta and several leading Chinese AI companies using a shared testing framework to identify potential safety and security issues.

Musk's reasoning is straightforward: companies should not be the only ones evaluating their own technology. Competitors may identify weaknesses, unexpected behaviors or security risks that an internal testing team could miss.

A New Layer of AI Safety
The proposal comes at a time when AI safety is receiving increased attention across the technology industry. Anthropic CEO Dario Amodei has also called for stronger safety measures, including independent evaluation of advanced AI systems and greater cooperation between AI companies.  

Independent or cross-company testing could provide an additional layer between AI development and public release. Instead of relying exclusively on a company's own evaluations, models could face external scrutiny from organizations with different testing methods and perspectives.

 

But There Are Challenges
The idea also raises practical questions. AI companies compete intensely with one another, and giving competitors access to advanced models could create concerns around intellectual property, confidentiality and competitive advantage. There would also need to be agreement on what should be tested, how results should be handled and what happens when a serious problem is discovered.

At present, Musk's proposal remains a proposal; the major AI companies have not agreed to participate in such a system.

The Bigger Picture
Whether this particular approach becomes reality or not, the discussion highlights an important shift in AI development: testing and safety are becoming part of the conversation alongside innovation and performance.

As AI becomes increasingly integrated into businesses, websites, applications and everyday digital services, the question is no longer simply “What can AI do?” It is also becoming “How do we make sure it does it safely and responsibly?”

The future of AI may therefore depend not only on building more capable models, but also on building better ways to test, evaluate and monitor them before and after they reach the real world.