Google NASDAQ:GOOG DeepMind, Google's artificial intelligence research unit, has proposed creating a new international watchdog to test and review the most advanced AI models before they reach the public. Chief Executive Officer Demis Hassabis said the organization could be staffed by independent technical experts, financed by leading AI companies and given up to 30 days to examine new models. The proposal quickly gained support from OpenAI, an artificial intelligence company led by Sam Altman, who described it as thoughtful, while Elon Musk called it a good starting point for wider discussions. Hassabis is expected to meet US policymakers in Washington next week as he attempts to turn the framework into a workable oversight system.
Hassabis said he decided to publish the plan after concerns emerged around the advanced cybersecurity capabilities of Mythos, a model developed by artificial intelligence company Anthropic. He described the model as a warning for society and suggested that increasingly powerful AI systems could also create biosecurity risks within the next couple of years. Google DeepMind has already been discussing its newest models with government officials and AI security institutes as those systems reach certain scores and benchmarks, but Hassabis argued that handling releases through an ad hoc process may not be sustainable. His proposed watchdog could introduce stronger measures if risks increase, including possible coordination among leading AI laboratories to slow development.
For investors, the proposal suggests that testing requirements and government scrutiny could become more important considerations for companies developing advanced AI systems. Anthropic and OpenAI have already delayed the broad release of their latest models under pressure from the Trump administration, leading to discussions between the European Union and the US over access to American AI technology. However, the plan may face political challenges because the US, EU and China have taken different approaches to AI regulation, while Congress has yet to approve meaningful federal legislation governing the technology. Hassabis believes there may now be enough momentum to establish a formal oversight structure, although he acknowledged that publishing the proposal is only the first step toward making it happen.