What if building a frontier AI model meant registering with a government watchdog, submitting to regular audits, and facing penalties for non-compliance, just like a publicly traded company? That is exactly the future Demis Hassabis, co-founder of Google DeepMind, is now advocating for. In interviews following the Trump administration's voluntary safety order, Hassabis proposed creating an independent oversight body modeled after the Securities and Exchange Commission or the Financial Industry Regulatory Authority, with real enforcement power over frontier AI development. His intervention reframes the regulatory debate at a moment when the industry is fractured between those who want no rules and those who want FDA-style pre-approval. For the thousands of founders building on top of or alongside frontier models, what Hassabis is proposing could become the template for how AI regulation actually looks in practice.
The SEC Model for AI: How Continuous Oversight Would Work
Hassabis's proposal envisions a dedicated independent regulator with authority to audit frontier AI models before they are deployed, require mandatory safety testing disclosures, and impose penalties for non-compliance. This is not a one-time approval process. Under the SEC-style model, frontier AI developers would be subject to ongoing regulatory scrutiny similar to what publicly traded companies face: regular reporting requirements, surprise audits, and enforcement actions for violations. The agency would have subpoena power, the ability to levy fines, and the authority to order changes to model deployment if safety concerns are identified. Unlike the FDA model, where a drug must be approved before it can be sold at all, the SEC model allows products to reach the market while the regulator monitors for systemic risk. The key difference is continuous oversight rather than a single gate. For startups building on frontier models, this means compliance would be a recurring cost built into operations, not a one-time hurdle to clear before launch.
Why Hassabis Rejects Both Voluntary and FDA-Style Approaches
Hassabis's positioning is significant because it stakes out a middle ground that rejects the two dominant camps in the AI regulation debate. On one side, the Trump administration's voluntary safety order asks labs to self-report and self-regulate with no enforcement mechanism. On the other side, the FDA for AI model championed by some safety advocates would require pre-market approval for any model above a certain capability threshold, effectively creating a government gatekeeper for AI releases. Hassabis argues that voluntary frameworks are insufficient for frontier AI systems that could pose existential risks, pointing to the financial industry's history as evidence that markets do not self-regulate when systemic risk is involved. But he also stops short of endorsing the FDA model, which he appears to view as too rigid and potentially innovation-stifling. The continuous oversight model he proposes would avoid the bottleneck of pre-market approval while still creating meaningful accountability. For founders, this Goldilocks framing makes Hassabis's proposal politically viable in a way that both extremes are not. If Congress ever passes AI legislation, the SEC model is the likeliest compromise.
What SEC-Style Compliance Would Cost AI Startups
If Hassabis's model becomes law, AI startups will face compliance costs that mirror what fintech companies have dealt with for a decade. SEC compliance for publicly traded companies typically costs 2 to 5 percent of revenue annually, covering legal fees, audit expenses, and reporting infrastructure. For AI startups, a similar regime would mean budgeting 5 to 10 percent of engineering time for compliance and reporting. This includes maintaining documentation of training data provenance, red-teaming results, deployment monitoring logs, and incident response procedures. The startups best positioned to survive this regulatory shift are those that build compliance infrastructure early. Just as fintech startups that invested in compliance from day one gained a competitive advantage when regulations tightened, AI startups that treat regulatory preparedness as a feature rather than a tax will have an easier path to scale. The compliance burden falls hardest on small teams. A 10-person startup can ill afford a dedicated compliance officer, but a 50-person startup that plans ahead can absorb the cost. This dynamic will favor startups that choose their regulatory strategy early and build around it.
The Political Calculus: Can This Model Actually Pass?
The timing of Hassabis's intervention is strategic. The Trump administration is finalizing its own approach to AI regulation, and the window for shaping that policy is narrowing. Multiple law firms and policy organizations are currently publishing analyses of the new executive order, and Hassabis's proposal injects a heavyweight voice into that conversation at a critical moment. The fact that a lab co-founder is publicly advocating for mandatory oversight is itself a signal: the frontier labs do not trust the voluntary model to work. This matters because it undermines the argument that self-regulation is sufficient. If the people building the most capable systems are saying they need a cop on the beat, the political case for legislation becomes much stronger. The key question is whether Congress can agree on the scope of such an agency. SEC-style regulation for AI would require defining what counts as a frontier model, setting capability thresholds, and determining jurisdictional boundaries with existing agencies like the FTC and NTIA. These definitional debates could stall legislation for years. But Hassabis's proposal provides a concrete model that lawmakers can debate, amend, and potentially pass. That alone moves the conversation forward.
What Founders Need to Do
Hassabis's proposal is not law yet, but it represents the most credible regulatory model on the table. Founders should treat the window before regulation as a grace period, not a permanent state. Three concrete steps to take now. First, start documenting everything: training data sources, model evaluations, safety testing results, and deployment monitoring logs. The cost of setting up documentation systems is minimal today and will be expensive to retrofit later. Second, budget for compliance. If a SEC-style watchdog arrives, expect 5 to 10 percent of engineering time to go toward regulatory reporting and audits. Startups that factor this into their hiring and roadmap planning will have a structural advantage. Third, watch the definitional debates. The single most important regulatory question for AI startups is how frontier model is defined. If the threshold is set too low, every API wrapper and fine-tuned model could face reporting requirements. If set too high, only the largest labs will be affected. Founders should engage in the policy conversation now, through trade groups, public comments, and direct advocacy. The regulation is coming. The only question is what form it takes, and founders who help shape that answer will fare better than those who wait to find out.