Trump plan to test AI models has a problem—US security teams were gutted by DOGE
Summary
Donald Trump signed an executive order (EO) expanding voluntary safety testing for frontier AI models, a move critics deem short-sighted and performative. The EO establishes no mandatory requirements for AI firms, instead setting up a voluntary collaboration process for safety reviews. An earlier, scrapped version of the EO sought 90 days of government access to models before other partners, but the signed version shortens this window to 30 days due to concerns about impeding AI innovation. The order directs the National Security Agency to create a classified benchmarking process for "covered frontier models" and, with the Treasury Department and CISA, establish a "cybersecurity clearinghouse." However, experts highlight the government's unpreparedness, noting the EO gives the Office of Personnel Management 60 days to expand cybersecurity hiring, while funding remains uncertain. Critics also point to significant cuts to CISA, which may hinder effective testing, and the challenge of defining "covered frontier models" given AI's emergent behaviors.
Key takeaway
For AI Security Engineers evaluating the impact of new government policies, Trump's voluntary AI safety testing order signals a need for proactive engagement. You should anticipate that government testing capacity will be limited, making your internal pre-release evaluations and transparency with federal partners even more critical. Focus on robust internal vulnerability detection and patching, as the EO's effectiveness hinges on industry collaboration and your willingness to share insights on emergent model behaviors.
Key insights
Trump's voluntary AI safety testing EO faces significant implementation challenges due to limited government capacity and reliance on industry goodwill.
Principles
- Effective AI defense requires finding and fixing vulnerabilities faster than adversaries.
- Frontier AI systems exhibit emergent behaviors, complicating fixed capability definitions.
- Government capacity for cybersecurity testing is crucial but currently diminished.
Method
The EO directs NSA to benchmark "covered frontier models," and NSA, Treasury, and CISA to establish a cybersecurity clearinghouse and voluntary submission framework.
In practice
- Prioritize enforcement against individuals using AI for illegal cyber activities.
- Continuously evaluate probabilistic, autonomous AI systems against evolving risks.
Topics
- AI Safety Testing
- Executive Order
- Cybersecurity Infrastructure Security Agency
- Frontier AI Models
- National Security Agency
- AI Regulation
Best for: CTO, VP of Engineering/Data, Director of AI/ML, Policy Maker, AI Security Engineer, Legal Professional
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.