Anthropic’s Recommendations to OSTP for the U.S. AI Action Plan

Source: Anthropic Date: 2025-03-06 URL: https://www.anthropic.com/news/anthropic-s-recommendations-ostp-u-s-ai-action-plan

Summary

Anthropic submitted recommendations to the White House OSTP for the US AI Action Plan, arguing “powerful AI systems will emerge in late 2026 or early 2027.” Six policy priorities: national security testing frameworks, tightened semiconductor export controls, enhanced lab security protocols, building 50 gigawatts of dedicated power capacity by 2027, accelerated government AI adoption, and economic impact preparedness mechanisms.

Implications

• Safety/policy posture thread. The “late 2026 or early 2027” timeline is the most specific frontier capability projection Anthropic has made publicly. If taken at face value, it implies the RSP’s ASL-3/4 thresholds could be reached within 18 months of this submission.
• 50 GW power by 2027. Advocating for 50 GW of dedicated AI power capacity is an infrastructure ask that benefits Anthropic directly — massive compute expansion is necessary for the training runs that frontier labs need. This is the same reasoning behind the Amazon 5GW energy announcement (April 2026).
• Export controls. Anthropic consistently advocates for stricter semiconductor export controls — this restricts Chinese AI development while constraining Anthropic’s competitors who might use offshore manufacturing. The policy overlap with Anthropic’s competitive interests is notable.
• Government AI adoption. Pushing accelerated government AI adoption is self-serving (Anthropic has multiple government contracts) but also strategically important — government as anchor customer validates frontier lab positioning.
• Watch: whether any of these six priorities made it into the actual US AI Action Plan; the “late 2026/early 2027” capability prediction against what actually ships.