OpenAI, Anthropic and Google DeepMind have been discussing the safety of the most advanced artificial intelligence systems for several weeks, without announcing a joint agreement. According to TechCrunch, these discussions could influence the voluntary practices of major labs, but their concrete scope remains unknown.
OpenAI confirmed to TechCrunch that the discussions concern the safety of the most powerful models and ways to cooperate between labs. The available information specifies neither the measures under consideration, nor the technical thresholds, nor the arrangements for sharing information.
Voluntary cooperation, without an announced framework
It is therefore impossible to know whether these talks will lead to operational commitments, joint assessments, or merely general principles. No protocol, timetable, shared criterion or external oversight mechanism is known, nor is any possible opening of the initiative to other players.
These discussions come as the Trump administration downplays concerns related to AI safety in the name of technological competition with China. The labs are thus seeking to reconcile accelerating model development with reducing the risks associated with their growing capabilities.
A signal for industry standards
This approach does not replace public regulation, but could serve as a framework for voluntary cooperation among companies developing the most advanced systems. If an agreement emerges, it could influence the industry's voluntary standards.
In France and Europe, where the AI Act already sets a legal framework for certain uses and risks, a private U.S. initiative would not directly alter regulatory obligations. It could nevertheless fuel debates on assessment methods, safety testing and the governance of advanced models.
Comments· 2 comments
The article makes this sound like a major breakthrough, but it barely addresses how meaningful a “united front” can be when these companies are still competing for investment, talent, and market share. I would have liked more scrutiny of what concrete commitments, if any, could come from these discussions rather than the familiar safety rhetoric.
That is a fair concern, but preliminary talks may still matter even if the companies remain competitors. I do not read the article as claiming that cooperation will solve everything; it suggests that shared risks could create at least some room for common standards.