On July 18, 2026, President Trump signed an executive order that replaces the most feared regulatory scenario for AI startups with a framework that looks dramatically different. The order establishes voluntary federal safety reviews for frontier AI models, abandoning the mandatory FDA-style pre-approval process that safety advocates and several Democratic lawmakers had championed. For AI founders who have been watching the regulatory landscape with growing anxiety, the shift is significant. But voluntary does not mean optional, and the order creates new dynamics that every founder needs to understand.
The executive order directs the Cybersecurity and Infrastructure Security Agency (CISA) and the Department of Homeland Security (DHS) to establish a framework for frontier AI model developers to voluntarily submit new models for security review before public release. The focus is on national security risks, cybersecurity vulnerabilities, and potential misuse. The order also mandates early government access to new AI releases for threat assessment, creation of AI safety testing best practices rather than binding standards, streamlined access for small banks and financial institutions to use AI models with federal guidance, and a cybersecurity clearinghouse for AI-related threats.
How the Industry Killed FDA-Style AI Regulation
The order represents a decisive victory for the industry faction that argued mandatory pre-approval would cripple US AI innovation. David Sacks, the venture capitalist and AI policy advisor with direct access to the Trump administration, led a sustained lobbying campaign arguing that requiring government sign-off before shipping an AI model would mirror the worst aspects of drug regulation, where getting a product to market takes years and costs billions. The argument that resonated most with the administration was the geopolitical one: forcing US AI companies into a lengthy pre-approval process would hand global AI leadership to China, which has no equivalent regulatory bottleneck.
The abandoned FDA for AI framework would have required mandatory pre-market approval for high-risk AI systems, modeled directly on the Food and Drug Administration's drug and medical device approval process. Under that framework, companies training frontier models above certain compute thresholds would have needed to submit safety documentation, pass government testing, and receive certification before releasing their models to the public. The cost and timeline implications for startups would have been severe. Industry estimates suggested that an FDA-style process could add 12 to 24 months and tens of millions of dollars to the launch timeline of any frontier model.
The Three Caveats Founders Cannot Afford to Miss
While the voluntary framework is a clear win for the industry, three structural dynamics mean that founders should not treat this as a regulatory free pass. First, voluntary frameworks in Washington rarely stay voluntary. Multiple law firms including Latham and Watkins, Mayer Brown, Skadden, and Davis Wright Tremaine have published analyses noting that companies should treat the framework as setting compliance expectations. When Congress eventually legislates on AI, the baseline is likely to be whatever the voluntary framework established. Companies that opted out of the voluntary process will face the most scrutiny.
Second, the early government access provision is a practical constraint on frontier model releases. Companies training models above certain compute thresholds must give CISA and DHS a look before going public. While this is not approval, it creates a de facto review period that could delay launches by weeks or months, depending on government bandwidth. The order does not define the compute threshold, but industry watchers expect it to target models at the GPT-4 scale and above, catching most frontier labs while leaving small startups and fine-tuned models unaffected.
Third, the small bank access provision creates a compliance-driven market opportunity. Banks and financial institutions that use AI models blessed through the voluntary federal review process will get streamlined regulatory guidance. This means AI model providers that can demonstrate CISA/DHS compliance will have a built-in sales channel to financial services customers who need regulatory clarity. Startups building for fintech and financial services should budget for the voluntary review process as a go-to-market requirement, not an optional checkbox.
What This Means for Non-US Founders
The order creates a two-tier system for non-US AI companies selling into the US market. Federally-reviewed models will get preferential access to US government customers and regulated industries like finance and healthcare. Foreign-developed models without a US federal review will face an implicit trust deficit, even if they meet their home country's regulatory standards. For international founders, the message is clear: if you want to sell to US enterprises or government agencies, plan to participate in the voluntary review framework.
The flip side is that the abandonment of FDA for AI removes what would have been a massive barrier to US market entry for non-US startups. Under the mandatory pre-approval model, any company selling into the US would have needed to pass the same costly approval process. The voluntary framework lowers that barrier considerably. For non-US founders, the calculus shifts from can we afford the regulatory process to do we want the compliance advantage that federal review provides.
The David Sacks influence angle is also instructive. The fact that industry advisors with direct ties to the administration successfully steered policy away from a mandatory framework demonstrates that regulatory outcomes are not predetermined. Founders who understand the policy process and have relationships in Washington can shape the rules that will govern their industry for the next decade.
