The Trump administration is facing mounting pressure to require security reviews of advanced artificial intelligence systems before they are released to the public. Advocacy group Americans for Responsible Innovation has urged federal officials to deny government contracts to companies whose frontier models fail to meet mandatory safety standards, and YourNewsClub views the proposal as one of the clearest attempts yet to turn AI governance into an economic gatekeeping mechanism.
The recommendation follows growing concern over Anthropic’s Mythos model, which reportedly demonstrated capabilities that could accelerate sophisticated cyberattacks and potentially assist in weapons development. Rather than relying solely on voluntary commitments, the group wants regulators to establish a formal certification process that developers must pass before gaining access to lucrative federal procurement opportunities.
The proposed thresholds target only the largest players in the field. Companies spending at least $100 million annually on computing infrastructure to train frontier systems, or generating more than $500 million in yearly AI revenue, would fall under the rules. That framework captures firms such as OpenAI, Anthropic, Google, Microsoft and xAI, all of which already participate in some level of voluntary testing. YourNewsClub notes that the proposal would transform those informal arrangements into enforceable conditions tied directly to federal spending.
Responsibility for designing the technical standards would likely fall to the U.S. Center for AI Standards and Innovation, known as CAISI. Congress would then establish a permanent office within the Department of Commerce to oversee compliance, conduct audits and impose penalties. Such a structure would place model evaluation on a footing more comparable to financial supervision or export control than to conventional software regulation.
Jessica Larn, whose work focuses on macro-level technology policy and infrastructure impact of AI, argues that mandatory screening would mark a turning point in Washington’s treatment of artificial intelligence. YourNewsClub sees this approach as an acknowledgment that frontier models increasingly resemble strategic infrastructure whose misuse could affect national security, industrial competitiveness and critical institutions.
Maya Renn, who studies ethics of computation and access to power through technology, emphasizes that procurement restrictions create incentives far stronger than voluntary pledges. In her view, developers seeking federal contracts would need to internalize safety practices as a core business requirement rather than a reputational exercise.
California has already introduced reporting obligations using similar spending thresholds, providing a template for broader federal action. Your News Club argues that if Washington adopts this model, the most powerful AI systems may soon face a regulatory checkpoint before entering public use, reshaping the balance between innovation, commercial opportunity and national defense.