Demis Hassabis, the CEO of Google DeepMind, thinks AGI might be only a few years away — and he wants the US to stop treating frontier AI safety as a voluntary exercise. His pitch: take today's self-regulated model testing and gradually turn it into a mandatory market-access regime. Initially, frontier AI companies would submit their models for review up to 30 days before release. Once the evaluation process matures, any model that crosses a certain capability threshold would have to pass a test before it could be deployed in the US. To oversee that, Hassabis proposes a new standards body modeled on FINRA, the US financial industry's self-regulatory watchdog — supervised by the federal government, primarily funded by AI companies, and staffed with independent experts plus representatives from the open-source community. The testing priorities? Cybersecurity, for starters.