NewsBite analysis · AI, Regulation, SecurityArs Technica

Trump plan to test AI models has a problem—US security teams were gutted by DOGE

Trump signed what his supporters describe as a balanced, safety-centered executive order to expand voluntary safety testing of frontier AI models. The text envisions the government coordinating with industry through a voluntary framework rather than imposing broad mandates, seeking to accelerate safe deployment while not stifling innovation. The administration intends to create a simplistic pathway for AI firms to submit models for safety testing, while the NSA would run a classified benchmarking process to determine when a model qualifies as a 'covered frontier model' and to scan for vulnerabilities at scale in collaboration with the Treasury and CISA. The order also directs the Office of Management and Budget to assess funding opportunities for vulnerability detection and to expand the federal cybersecurity workforce. Critics, however, say the approach is performative and underpowered: it relies on voluntary cooperation, provides no firm requirements for firms, and relies on a government with reported staffing shortages and past contract slowdowns. The differences between the leaked draft and the final text include a shortened 30-day testing window compared with an earlier 90-day window, a move widely seen as an attempt to preserve U.S. AI leadership at the expense of safety oversight. Experts from CFR warn that defining what counts as a 'covered frontier model' and ensuring real observability will be the biggest challenges, and they caution that even with a formal process, meaningful testing is hard to implement quickly in a fast-evolving field. The piece underscores that the outcome will depend on how transparently the program operates and whether industry partners participate in good faith.