OpenAI is working with Anthropic and Google DeepMind to explore greater coordination around artificial intelligence safety, according to Reuters.
The companies have reportedly discussed creating an industry standards organization that could establish common expectations for the development and testing of advanced AI models.
The talks are notable because the three organizations are among the most important competitors in frontier artificial intelligence. Each is investing heavily in models, computing infrastructure, enterprise products, and consumer applications.
Competition creates pressure to develop and release increasingly capable systems quickly. A shared standards framework could reduce the risk that one company lowers its safeguards to gain a commercial advantage.
Potential areas of cooperation could include independent evaluations, cybersecurity testing, disclosure of dangerous capabilities and procedures for responding to models that demonstrate unexpected behavior.
OpenAI global policy chief Chris Lehane said the companies had been discussing safety coordination for several weeks. The discussions remain preliminary, and no final organizational structure has been announced.
The participants are considering whether cooperation could proceed without a special antitrust exemption. Companies that coordinate industry practices must avoid agreements that could improperly restrict competition, divide markets or control prices.
A carefully structured standards body could focus on technical safety requirements while leaving the participants free to compete on products, performance, pricing and distribution.
The discussions follow increasingly urgent warnings from executives at major AI laboratories. Leaders have raised concerns about models becoming capable of conducting sophisticated cyberattacks, assisting with biological threats or operating digital tools with limited human oversight.
At the same time, the industry remains divided over whether slowing development is practical or desirable. Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg have argued that individual companies can improve safety without coordinating a broad slowdown.
The disagreement highlights the challenge of developing common standards across organizations with different business models. OpenAI and Anthropic develop largely proprietary frontier models, while Google combines model development with cloud services, advertising and consumer products.
International competition adds another complication. U.S. companies may be reluctant to slow development if they believe developers in China or other countries will continue advancing without equivalent restrictions.
A standards organization would therefore need to demonstrate that safety requirements can reduce serious risks without preventing useful innovation or weakening the competitiveness of participating companies.
The initiative could also influence government regulation. Policymakers may be more willing to allow industry flexibility if leading developers establish credible testing, reporting and accountability systems.
However, voluntary standards are likely to face questions about enforcement. An organization controlled by the companies it oversees would need transparent procedures and meaningful independent participation to build public confidence.
The discussions represent an early attempt to create shared safeguards before increasingly autonomous systems become widely deployed. Whether the companies can translate broad agreement about safety into specific, enforceable standards remains the central question.
