Trump AI testing plan faces problem: DOGE gutted US security teams

This Ars Technica article focuses on the institutional contradictions the Trump administration faces in pushing forward a plan for AI model safety testing. The report notes that the White House had discussed using an executive order to require the government to conduct safety testing before advanced AI systems are deployed to federal, state, or local government, with the goal of reducing national-security, cyberattack, and large-scale public-safety risks. On the surface, this represents the Trump administration, after rescinding the Biden-era AI executive order, once again acknowledging the importance of AI safety testing; but critics argue that this policy may be short-sighted, highly performative, and lacking a sufficient foundation for execution. The problem is that DOGE's cuts to government departments and tech-security teams have already weakened the professional capacity the United States originally had to conduct such testing. The article also mentions that government bodies with AI safety-testing expertise, such as CAISI, were originally regarded as important units that could jointly evaluate national-security and public-safety risks alongside companies; but if the relevant personnel and resources are cut, then even if the government announces it will test AI models, it may be unable to carry out in-depth, sustained, independent review. Another point of contention is that the model designs of many large AI companies are opaque, making it difficult for outside testers to grasp the full range of risk sources. If the government wants on the one hand to avoid over-regulating companies and maintain so-called light-touch regulation, while on the other hand hoping to use testing to boost public trust, critics question whether these two goals are necessarily compatible. For developers, researchers, and policy observers, what this article warns about is not the risk of a single model, but the capability of AI governance itself: without a stable safety agency, professional talent, testing authority, and transparency, an executive order may merely add a procedural formality ahead of dangerous deployment, rather than a substantive line of defense.