OpenAI and Anthropic abandoned a deal to mutually test each other's AI models, despite reaching the contract stage. The collapse highlights the challenges of voluntary safety agreements in the AI industry, with significant implications for Microsoft, a major investor in both companies.
TL;DR
- OpenAI and Anthropic abandoned a deal to test each other's AI models, despite reaching the contract stage.
- The collapse highlights the challenges of voluntary safety agreements in the AI industry.
- Microsoft, a major investor in both companies, faces increased risk due to the abandoned deal.
What happened
OpenAI and Anthropic reached the formal contracting stage on an agreement to run mutual safety tests on each other's models. However, the deal was abandoned for unknown reasons, according to reporting summarized on The AI Daily Brief.
The idea of cross-testing has gained prominent endorsements, with Elon Musk and Mustafa Suleyman, Microsoft's AI chief, calling it a smart safety idea. However, the commercial and legal complexities have proven to be significant barriers.
Microsoft holds a roughly 27% stake in OpenAI, valued at about $135 billion, and booked a $3.2 billion gain on its Anthropic investment in fiscal Q4 2026. The collapse of the deal raises risks for Microsoft's investments and cloud contracts.
Why it matters
The abandonment of the deal highlights the structural challenges of voluntary safety agreements in the AI industry. Giving a rival access to probe your model involves revealing trade secrets and potential vulnerabilities, which can have immediate commercial costs.
The safety benefit of such agreements is diffuse and long-term, making it difficult to justify the immediate costs. This asymmetry is why voluntary cross-testing may be structurally unstable.
For Microsoft, the collapse of the deal raises risks as it is a major investor in both OpenAI and Anthropic. The company's exposure to these frontier labs is significant, and the lack of mutual safety testing could impact Microsoft's balance sheet and cloud contracts.
Key facts
- OpenAI and Anthropic reached the formal contracting stage on an agreement to run mutual safety tests on each other's models.
- The deal was abandoned for unknown reasons, according to reporting summarized on The AI Daily Brief.
- Microsoft holds a roughly 27% stake in OpenAI, valued at about $135 billion.
- Microsoft booked a $3.2 billion gain on its Anthropic investment in fiscal Q4 2026.
- Elon Musk and Mustafa Suleyman, Microsoft's AI chief, have endorsed the idea of cross-testing.
- Microsoft's exposure to OpenAI and Anthropic raises risks for its balance sheet and cloud contracts.
- The commercial cost of revealing model vulnerabilities is immediate, while the safety benefit is diffuse.
- Voluntary cross-testing may be structurally unstable due to the asymmetry in costs and benefits.
Context
The AI industry is facing increasing scrutiny over safety and ethics. Voluntary agreements between companies to test each other's models have been proposed as a way to improve safety, but the commercial and legal complexities have proven to be significant barriers.
Microsoft's investments in OpenAI and Anthropic highlight the risks and opportunities for major tech companies in the AI industry. As a major investor, Microsoft has a significant stake in the success and safety of these frontier labs.
The collapse of the deal between OpenAI and Anthropic underscores the need for regulatory frameworks to ensure the safety and ethical use of AI. Without such frameworks, voluntary agreements may not be sufficient to address the risks and challenges of the AI industry.
