Anthropic submits AI Action Plan recommendations to White House OSTP
The submission urged tighter H20-chip export controls, classified channels between labs and intelligence agencies, and 50 gigawatts of new US power capacity by 2027.
- Government & policy
- Notable
Anthropic submitted formal recommendations to the White House Office of Science and Technology Policy as part of the Trump administration’s request for input on a national AI Action Plan. The submission set out a series of specific policy asks rather than general principles, and was framed around capabilities the company expected AI systems to reach by late 2026 or early 2027 — described in the submission as comparable to Nobel-laureate-level expertise across disciplines.
The recommendations covered several areas. On national security, Anthropic called for government agencies to build the capacity to test both domestic and foreign AI models for security risks, and for standardised evaluation frameworks to do so. It urged tighter semiconductor export controls, specifically naming restrictions on Nvidia’s China-market H20 chips and proposing that large chip deployments require government-to-government agreements. It called for classified communication channels between AI labs and US intelligence agencies, expedited security clearances for industry researchers, and new security standards for frontier labs generally. On infrastructure, it recommended a target of 50 additional gigawatts of dedicated US power capacity by 2027, alongside streamlined permitting to enable it. It also urged government agencies to inventory workflows that could benefit from AI and to modernise economic data collection ahead of anticipated labour-market disruption.
The submission was one of several the administration solicited from labs, companies and advocacy groups as it prepared the plan, and it positioned Anthropic — which had built its public identity substantially around AI-safety research — as pushing not for caps or moratoria but for testing infrastructure, export controls and energy buildout, treating rapid capability growth as close to certain and the appropriate response as anticipatory infrastructure rather than a slowdown. It appeared the same week as Forethought’s “Preparing for the Intelligence Explosion” essay, which argued from a similar premise about the pace of anticipated progress toward the conclusion that governance needed to be worked out well in advance.