White House Approves $9B for CIA/NSA AI Chips; Trump Delays AI Oversight Order

The U.S. government approved $9 billion in classified AI chip spending for spy agencies while shelving a plan to give the NSA oversight power over frontier AI models.

🏛️ White House Approves Secret $9 Billion AI Chip Buy for CIA and NSA

Decoded: The White House approved a classified $9 billion request for advanced AI chips destined for the CIA, NSA, and other intelligence agencies, the New York Times reported on May 22. Agency officials warned that the intelligence community cannot fully deploy the latest AI models on its classified networks because of a shortage of cutting-edge accelerators — a gap that limits real-time AI capabilities on air-gapped systems where commercial cloud access is not permitted. The funding is part of a classified budget supplemental. Separately, Politico reported on May 20 that NSA and Cyber Command launched a task force codenamed Mythos to determine how AI models from OpenAI and Google (GOOGL) can be safely deployed across classified national security missions. (New York Times, May 22, 2026; Politico, May 20, 2026)

Why it matters: A $9 billion government AI chip order is a demand vector analysts have not fully priced into supply forecasts. Intelligence community procurement operates on classified acquisition timelines that bypass normal commercial queues — making this spend additive to hyperscaler demand already driving shortages. The NSA Mythos task force signals the intelligence community is moving from AI evaluation to active deployment, requiring sustained hardware procurement well beyond the initial $9 billion. Palantir (PLTR), which holds the most classified government AI contracts of any publicly traded company, is positioned to expand its AIP platform deployments as agencies acquire the compute to run models at scale. Government demand is now a third pillar alongside hyperscalers and enterprise — less visible but potentially more durable, since national security budgets do not face the ROI scrutiny of corporate AI spending.

🏛️ Trump Shelves AI Model Oversight Order After Last-Minute Objections

Decoded: President Trump postponed signing an executive order that would have granted the federal government authority to evaluate frontier AI models before public release, the New York Times reported on May 21. Trump cited concerns about "aspects of" the order. Under the drafted text, the NSA would have served as lead agency for final safety determinations, with voluntary testing frameworks applied to models from companies including Anthropic and Google (GOOGL). Politico reported the order was specifically triggered by concerns over Anthropic Mythos — described as a powerful, cyber-focused AI with dual-use national security implications. The administration has not announced a revised timeline. (New York Times, May 21, 2026; Politico, May 20, 2026)

Why it matters: Trump pullback removes a near-term regulatory risk for frontier AI labs, particularly Anthropic, whose Mythos model was the stated trigger for the order. For OpenAI, which is preparing a confidential IPO filing targeting September, the absence of a mandatory pre-release evaluation regime reduces regulatory overhang at a critical pre-IPO window. The episode exposes a structural tension inside the administration: national security officials want authority over dual-use AI models while industry advisors with equity stakes pushed back on mandatory oversight. The delay is not a permanent retreat — pre-release AI evaluation is likely to return in a narrower or voluntary form that carries less liability for developers.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team