Cybersecurity Policy Report, Senate Bill Would Require Prerelease Testing of Cyber-Capable AI Models, (Jul 21, 2026)
Developers of “frontier” AI (artificial intelligence) models that can exploit cybersecurity vulnerabilities would have to submit their models to the National Security Agency prior to their public release under legislation unveiled today by Sen. Mark Warner (D., Va.), vice chairman of the Senate Intelligence Committee.
The Secure AI Development Act (S.B. __, 119th Cong. (2026)) “would establish a mandatory secure testing environment for the nation’s most advanced AI models before deployment, modernize the federal government’s processes for identifying and disclosing AI-related cybersecurity vulnerabilities, improve information sharing between government and AI developers, and create a voluntary AI safety incident reporting system,” Sen. Warner explained in a news release.
“There are artificial intelligence models that are emerging and are capable of autonomously identifying and exploiting severe software vulnerabilities,” Sen. Warner noted in an AI legislative framework released today.
“These AI models are capable of working faster than humans at these tasks, increasing cyber-attack vectors and requiring significant resources to defend against. Currently, these capabilities are believed to be confined to a few known models, but as model capabilities increase, there must be a statutory structure in place purpose-built to address the need to determine a given model’s capabilities prior to the model’s public release,” the roadmap says.
Under the Secure AI Development Act, frontier models would have to be registered with the National Institute of Standards and Technology and submitted to an NSA-led Artificial Intelligence Security Center 21 days before their public release, according to the bill’s text.
The bill would direct the NSA to “establish a process by which critical infrastructure operators, as well private sector entities that develop or maintain information systems utilized by critical infrastructure operators, shall access a secure test-bed for the purpose of testing and evaluating the impact of frontier artificial intelligence models on information systems maintained by critical infrastructure operators prior to public release or distribution of such models,” it says.
An executive order signed by President Trump last month encourages—but doesn’t require—developers of frontier AI models to submit their models for testing 30 days prior to public release (CPR, June 3). But the administration also has used its export control authority to block the release of at least one model developed by Anthropic (CPR, June 15).
The Secure AI Development Act is among several bills unveiled today as part of Sen. Warner’s AI legislative roadmap to address the impact of AI on society, including workforce disruptions, the proliferation of data centers needed to power AI, and the use of AI to generate child sexual abuse material or non-consensual intimate images.
The legislative package includes the Artificial Intelligence Access, Gatekeeper Exchange, and Nondiscriminatory Transfer (AI AGENT) Act, which Sen. Warner introduced in draft form last month (CPR, June 30).
“Artificial intelligence will reshape nearly every aspect of our economy and society. The only question is whether Congress is going to help shape that future or spend the next decade scrambling to catch up,” Sen. Warner said.
“America doesn’t have to choose between leading in AI and putting commonsense rules of the road in place,” he added. “In fact, our leadership depends on getting this right. We need policies that encourage innovation while protecting consumers, strengthening competition, preparing our workforce, and ensuring AI advances America’s national security, not undermines it.”
MainStory: TopStory FederalLegislation AINews DataSecurity DataPrivacy