Google DeepMind CEO Demis Hassabis has endorsed the broad direction of an AI safety proposal put forward by Anthropic CEO Dario Amodei, adding another major lab leader to calls for stronger safeguards around frontier models. In a post on X, Hassabis said Amodei’s essay points in the right direction, though he stopped short of backing every element of the plan.
Amodei’s proposal calls for controlled slowing of frontier-model progress where necessary, independent evaluators with access to AI laboratories, and shared safety standards across companies. The aim is not to halt AI development, but to give safety measures a chance to keep pace with increasingly capable systems.
- A controlled approach to advancing frontier AI models rather than an outright development freeze
- Independent external evaluators able to assess work inside AI labs
- Common industry standards for the safety and governance of frontier models
Hassabis’s position remains more cautious than Amodei’s. He has not signed on to a fully defined mechanism, but connected his support for the proposal’s overall direction to Google DeepMind’s own push for an industry-level standards body focused on frontier AI.
The comments follow reported support from OpenAI CEO Sam Altman for the principle of outside evaluators. Agreement on parts of the safety agenda among the leaders of Anthropic, OpenAI and Google DeepMind is notable because the companies are also competing for model performance, enterprise customers, developers and compute infrastructure.
There is no joint agreement yet, nor an industry-wide limit on the pace of AI development. The key test will be whether calls for independent scrutiny and common standards turn into commitments that still hold when they affect release schedules, market share or the race to deploy more capable models.
