OpenAI says it has been working with Anthropic and Google for several weeks on AI safety, and it does not need an antitrust waiver to coordinate on safety
Maggie Eastland /Bloomberg Law:
Context & Ripple Effects
OpenAI and Anthropic had already given the US AI Safety Institute early access to major models in 2024, placing external evaluation alongside their internal safety processes. In 2025, the two labs also published joint tests of each other’s models to identify blind spots.
Reports of working-group meetings on an industry-led standards body add Google to that collaboration. OpenAI’s position that safety coordination does not require an antitrust waiver frames the effort as work the companies can continue under existing competition rules.
First-order effects
- OpenAI, Anthropic, and Google can continue their safety discussions without waiting for a bespoke antitrust exemption, according to OpenAI’s stated legal view.
- The three companies gain a shared venue for translating cross-lab safety testing into potential industry standards.
Second-order effects
- Other AI developers face pressure to engage with standards shaped by OpenAI, Anthropic, and Google or explain why their safety practices differ.
- The US AI Safety Institute’s earlier-access arrangement becomes complementary to, rather than a replacement for, private coordination among major model developers.
Third-order effects
- If an industry-led body takes shape, a small group of frontier-model providers may increasingly set the operational baseline for AI safety before formal rules are established.
- The pattern points to AI governance combining government evaluation with voluntary coordination among the companies building leading models, raising the importance of who participates in those forums.
The trend: AI safety governance is moving toward hybrid oversight: public model evaluation paired with cross-company standards-setting by major AI labs.