Hassabis Calls for US-Led AI Watchdog, Targets 2026 Launch as AGI Looms

For AI practitioners and researchers, Hassabis' proposal strikes at the heart of model deployment lifecycles. The envisioned watchdog would introduce mandatory pre-release evaluations and potentially halt rollouts — a direct challenge to the 'move fast' culture. As AGI timelines compress, the plan demands that the AI community rethink not just model safety but the governance architecture that will shape their work.

Demis Hassabis, CEO and co-founder of Google DeepMind, has ignited the AI governance debate by proposing a new US-led international watchdog to oversee frontier artificial intelligence models before they are deployed. In a manifesto titled "A Framework for Frontier AI and the Dawning of a New Age," published on July 14, 2026, Hassabis argues that the rapid approach of artificial general intelligence (AGI) — which he says is "probably only a few short years away" — demands a systemic oversight body with real teeth. The proposal calls for an industry-funded organization, modeled loosely on the Financial Industry Regulatory Authority (FINRA), staffed by technical experts and operating under US government supervision. Its mandate would include evaluating the most advanced AI systems before public release and coordinating an industry-wide pause if risks are deemed too high. The watchdog would cover both proprietary frontier models and qualifying open-source systems, reflecting a nuanced approach to the open versus closed model debate.

“

Demis Hassabis, CEO and co-founder of Google DeepMind, has ignited the AI governance debate by proposing a new US-led international watchdog to oversee frontier artificial intelligence models before they are deployed.

The timing of Hassabis' push is no coincidence. It comes amid heightened geopolitical tension between the US and China over AI supremacy and recent government interventions in advanced model releases. Cybersecurity risks from current AI systems, Hassabis warns, are merely an "early warning" of more grave threats — including those in biological and nuclear domains — that future frontier AI could enable. By embedding the watchdog under US oversight while involving international experts, the proposal aims to set global standards anchored in American leadership, a posture that may appeal to the Trump administration, with which Hassabis has quietly consulted for months. The executive revealed that feedback from the administration has been "very positive," and he aims to have the organization operational before the end of 2026.

For the AI industry, this plan marks a potential shift from voluntary commitments to enforceable pre-deployment reviews. Currently, there is no comprehensive national AI law in the US and no binding global treaty on AI. The proposal could reshape competitive dynamics: companies would face a common evaluation threshold, potentially leveling the playing field between giants like Google, OpenAI, and emerging players. However, the reliance on industry funding and US leadership raises concerns about impartiality and geopolitical bias. European and Chinese regulators might view a US-dominated body with skepticism, especially given the bloc's own efforts with the EU AI Act. Moreover, the inclusion of open-source models — a fiercely debated topic — signals that even decentralized development is on the regulatory radar.