Four of the leading artificial intelligence labs have begun publicly advocating a reduction in the pace of development of the most advanced models. The position gained momentum this weekend after Anthropic CEO Dario Amodei proposed new safety restrictions, with support from Sam Altman, Elon Musk and Demis Hassabis. There is still no formal agreement among the companies.
Amodei said the models' capabilities are advancing faster than the mechanisms used to understand and control them. Among the concerns is the growing use of AI itself in the development of more advanced systems, which could accelerate this process even further.
Anthropic and OpenAI will expand external evaluations
The proposal calls for independent teams to have continuous access to the labs' systems to evaluate models, investigate incidents and verify safety practices. Anthropic committed to adopting this model, and Sam Altman said that OpenAI will do the same.
The discussion gained momentum after an incident in cybersecurity tests in which OpenAI models managed to bypass controls and access external systems. The company itself described the episode as a warning sign for the advance of more autonomous agents.
Elon Musk, of xAI, supported Amodei's proposal. Demis Hassabis, CEO of Google DeepMind, also said the initiative is heading in the right direction, although it is still necessary to define how any common rules would be applied.
Coordinated slowdown still has no formal agreement
The statements do not mean a halt in AI development. The proposal is to make new advances conditional on more robust evaluation and safety mechanisms, in addition to creating shared standards among the leading labs.
For now, the concrete result is the adoption of external evaluations by Anthropic and OpenAI. A coordinated slowdown among companies would still depend on common rules, independent oversight and regulatory issues, including possible antitrust restrictions.
