Trump Orders Immediate Halt to Anthropic AI Across Federal Agencies, Six-Month Wind-Down Begins
Feb 28, 2026 - President Donald Trump directed all federal agencies, including the Department of Defense, to stop using Anthropic's AI tools effective immediately, with a six-month period to unwind existing deployments. The order follows a public standoff over use restrictions that Anthropic refused to remove for defense applications.
What the directive requires
Agencies must cease use of Claude and other Anthropic products now and plan full removal within six months. Trump said on social media that a private company will not "dictate" how the military operates, and Defense Secretary Pete Hegseth emphasized the department's position of "full, unrestricted access" for any lawful defense purpose.
Why this happened
Talks between the Pentagon and Anthropic broke down after the company declined to permit unrestricted military use of its models. CEO Dario Amodei said Anthropic could not agree to terms that might enable mass surveillance of Americans or fully autonomous weapons, and argued proposed "compromise" language would let those safeguards be bypassed.
Officials warned Anthropic could be labeled a supply chain risk and suggested the government could use the Defense Production Act if needed. For context, the Defense Production Act lets the government prioritize and allocate industrial capacity for national security needs. See background at the Congressional Research Service's summary of the law here.
Reactions across government and industry
Hegseth called Anthropic's stance a "master class in arrogance," while Pentagon spokespeople rejected claims they intend to use AI for illegal surveillance or autonomous weapons. Amodei countered that some AI uses remain unsafe and could undermine democratic values.
The industry split is unusual: Elon Musk sided with the administration, while OpenAI CEO Sam Altman publicly backed Anthropic's red lines and questioned the Pentagon's "threatening" approach. Current and former defense officials, including retired Gen. Jack Shanahan, signaled sympathy for narrow safeguards and noted Claude is already in use across parts of government, including classified settings.
Who's affected
This touches any federal program using Anthropic models directly or indirectly (prime contracts, subcontracts, platform integrations, cloud marketplaces, and pilot projects). If "supply chain risk" designation proceeds, it could extend beyond direct licenses to partners and integrators that embed Anthropic models.
Immediate steps for agency leaders, CIOs, and program managers
- Freeze usage: Halt production and new deployments of Anthropic tools. Disable API keys, block endpoints at the network layer, and suspend single sign-on access where applicable.
- Inventory exposure: Identify all systems, pilots, and prototypes using Claude or Anthropic APIs (including contractor-operated systems). Map data flows, model endpoints, and business owners.
- Data hygiene: Export permitted data, rotate secrets, and purge cached prompts/responses in line with records schedules and security policies. Confirm deletion or retention per agency SOPs.
- ATO and security: Update system security plans, POA&Ms, and ATO documentation to reflect removal plans. Re-run impact assessments for any downstream systems relying on Anthropic outputs.
- Contracts and funding: Notify COs and CORs; initiate terminations for convenience or no-cost modifications as appropriate. Address CLINs, ceilings, and deliverables tied to Anthropic usage.
- Continuity of operations: Stand up a transition team. Define minimum viable alternatives for mission tasks (search, drafting, translation, code assistance, analysis) to avoid service disruption.
- Alternatives assessment: Evaluate compliant providers and open-source options against mission needs, data controls, logging, latency, and export restrictions. Perform security and privacy due diligence before redeployment.
- Communications: Brief leadership, IG, OGC, workforce, and affected partners. Publish internal FAQs with timelines, approved substitutes, and points of contact.
- Risk posture: Track any "supply chain risk" designation and adjust SBOMs, SSPs, and vendor risk registers. Prepare for additional interdiction or interdiction-like measures if announced.
Procurement and policy notes
The core dispute centers on whether a vendor can place usage limits beyond "lawful purpose" in national defense contexts. Agencies should ensure future solicitations specify acceptable safeguard boundaries, rights to operate models, and contingency clauses for rapid provider substitution.
If the Defense Production Act is invoked, acquisition teams may face priority ratings and allocation requirements. Coordinate early with legal counsel, component acquisition executives, and OMB for unified guidance.
Market implications to monitor
- Whether DoD formally designates Anthropic a supply chain risk, and the scope of that designation.
- How other leading providers respond to similar requests, which could affect interoperability and multi-model strategies.
- OMB, CISA, and DoD CIO guidance on standardized transition playbooks and approved alternatives.
- Impacts to ongoing classified programs and cross-domain solutions that integrated Claude.
Helpful references
- DoD AI Ethical Principles overview: defense.gov
- AI for Government - practical guidance on public-sector AI adoption, governance, and vendor transitions.
Bottom line: Stop usage now, plan for full removal within six months, document the transition, and secure mission continuity with vetted alternatives. Keep leadership looped in, tighten controls on data and access, and get ahead of any supply chain directives that may follow.
Your membership also unlocks:
