OpenAI and Anthropic reportedly to exchange API access for safety

Technology Sep 21, 2026

OpenAI and Anthropic are reportedly in talks to let each other test their AI models for safety.

The idea is simple: by sharing API access, both companies can spot risks or weird behavior that might slip through their own checks.

Importantly, neither side gets to keep any data from these tests.

Collaborative testing could catch reward hacking

With AI getting more complex and sometimes unpredictable, this move could help catch problems early, like reward hacking or unexpected actions.

It's a rare example of big tech rivals working together for safer AI, echoing calls from leaders like Sam Altman and Dario Amodei for better oversight in the industry.