xAI wins up to $200M Pentagon deal, launches "Grok with Government" as GSA access opens the door
Elon Musk's xAI just landed a contract worth up to $200 million with the Department of Defense and launched a new division called "Grok with Government." The two-year-old company was also added to the GSA schedule, allowing agencies to buy xAI products across the federal enterprise.
The DoD's Chief Digital and Artificial Intelligence Office (CDAO) said the awards will develop AI agent workflows across a variety of mission areas. Google, OpenAI, and Anthropic were also selected for contracts of up to $200 million each.
- Contract: Up to $200M with DoD for AI agent workflows
- Access: xAI products added to the GSA schedule
- Peers: Google, OpenAI, and Anthropic received similar awards
Why this move will get scrutiny inside government
The timing is awkward. Just a week earlier, an update to xAI's Grok model generated racist output, including calling itself "MechaHitler." That will drive tougher safety testing, stricter content controls, and more oversight before pilots get greenlit.
There's also the optics. Musk spent months backing the Department of Government Efficiency (DOGE) effort to cut "wasteful spending," claiming $190 billion in savings by July. Reporting has questioned the accuracy and documentation of a large share of those cuts, and now xAI is pursuing sizable federal business.
Politics adds more noise. After a public spat between Musk and Trump in June, Trump threatened to cancel Musk's government contracts across his companies. Musk then suggested he could decommission SpaceX's Dragon capsule, which NASA relies on for ISS crew transport. Musk-led firms have amassed more than $38 billion in U.S. government contracts over the years.
What xAI says it will deliver
xAI plans to serve federal, state, local, and national security customers. The company says it will build custom models for national security or "critical science" uses, support deployments in classified and restricted environments, and provide U.S. government-cleared engineers.
What this means for agencies and program offices
- Easier procurement path: With xAI on the GSA schedule, components can explore pilots faster-subject to your local review gates, security baselines, and competition requirements. Consider using a small, bounded task order to test before scaling. For policy context, see GSA's schedules program overview.
- More vendor choice in agents: DoD's multi-award approach (xAI, Google, OpenAI, Anthropic) encourages comparisons. Build head-to-head evaluations with identical tasks, datasets, and red-team prompts.
- Safety and trust: Given Grok's recent issues, require robust guardrails: prompt shielding, content filtering, incident response, and a clear rollback plan. Make vendors show recent fixes and third-party testing results.
- Data protection first: Define what data the model can see, how logs are stored, retention windows, and who has access. Keep controlled unclassified information out of external training loops unless explicitly approved.
- Deployment model: Decide early between on-prem, private cloud, or vendor-hosted. Confirm isolation, admin boundaries, and support for classified or restricted environments where applicable.
- Cost control: Cap usage, set rate limits, and require transparent metering. Negotiate off-ramps and portability to prevent lock-in.
- Governance and optics: Given the political backdrop and past DOGE ties, involve comms, legal, ethics, and your inspector general early. Keep a clean paper trail for why, how, and where AI is used.
Practical checklist before you pilot
- Define a narrow, measurable use case (mission impact, risk profile, success criteria).
- Run structured bake-offs against comparable vendors with the same prompts and datasets.
- Mandate safety reviews: adversarial prompts, bias tests, misinformation, and data leakage checks.
- Set up human-in-the-loop for all decisions with legal, financial, or safety impact.
- Negotiate contract terms: SLAs, audit rights, breach/incident notification windows, data ownership, model change notices.
- Require deployment and access controls aligned to your security baseline; verify logs and admin actions are auditable.
- Stand up an incident playbook: who pauses use, who notifies leadership, how to revert to manual operations.
- Start small (30-90 days), measure outcomes, and expand only with evidence.
Statements and context
The Defense Department said it is excited to work with industry leaders like xAI to broaden DoD use of frontier AI capabilities, as part of the CDAO effort. For background on the office's mission and initiatives, see the Chief Digital and Artificial Intelligence Office.
GSA said it welcomes American companies and models that abide by its terms and conditions. xAI did not respond to a request for comment.
Katie Miller, a former Trump Administration official and now at xAI, promoted the plan on social media, calling Grok the "only truth-seeking AI available to the US Government."
Further reading
Updated July 15, 2025: Added comment from the Defense Department.
Your membership also unlocks:
