Trump Signs Executive Order To Review Ai Models Before They’re Released

President Donald Trump signed an executive order Tuesday creating a “voluntary framework” for AI companies to stock their frontier models pinch nan national authorities earlier they’re released “to beforehand unafraid invention and fortify nan cybersecurity of captious infrastructure.”

The bid says nan US AI manufacture has succeeded successful portion “because we garbage to stifle this invention pinch overly burdensome regulation,” but that it besides recognizes caller AI capabilities travel pinch information risks. Accordingly, it directs respective national agencies to travel up pinch a model to “assess nan precocious cyber capabilities of AI models” earlier they’re released to nan public. Companies would person nan discretion of whether to stock their models pinch nan authorities pre-release, but could get definite confidentiality protections if they take to do so. It besides requires nan national authorities to hole cyber defenses for AI, particularly for captious infrastructure.

The bid comes aft Trump postponed astatine nan past minute signing a antecedently planned executive bid that he worried could “get successful nan way” of competing pinch China. While nan earlier type would person allowed AI companies to voluntarily stock their models 14 to 90 days earlier release, according to The New York Times, nan existent type asks companies to stock their models up to 30 days earlier nationalist release. Google, Microsoft, and xAI agreed past month to let pre-release reappraisal by nan Commerce Department’s Center for AI Standards and Innovation (CAISI).

OpenAI and Anthropic had already agreed to stock their models pinch CAISI backmost successful 2024 nether President Joe Biden arsenic portion of Biden’s push for AI information guardrails. But until recently, nan Trump management downplayed information concerns and took a hands-off attack under erstwhile White House AI czar David Sacks. The bid signed Tuesday explicitly says it shouldn’t beryllium taken arsenic a shape of mandatory licensing aliases preclearance. Still, it reflects immoderate willingness of nan Trump management to employment oversight of AI companies.

One facet successful this displacement whitethorn beryllium Anthropic’s constricted April rollout of its powerful Mythos model, which nan institution said had flagged “thousands of high-severity vulnerabilities, including immoderate successful each awesome operating strategy and web browser.” Mythos besides seemed to create an opening to thaw tensions betwixt Anthropic and nan administration, pursuing its legal conflict pinch nan Pentagon complete its usage of AI for autonomous lethal weapons and wide surveillance.

The recently signed bid has truthful acold garnered praise moreover from groups that person advocated against restrictions connected authorities AI laws. “The White House is officially Mythos-pilled,” Americans for Responsible Innovation President Brad Carson said successful a statement, adding that nan bid shows nan Trump management is taking AI vulnerabilities seriously. Alliance for Secure AI CEO Brendan Steinhauser said his group is “pleased to spot that nan Trump management is taking nan risks of these models seriously,” and some Steinhauser and Carson urged Congress to codify mandatory protections.