DoD awards up to $200M AI contracts to xAI, Anthropic, Google, and OpenAI
The Pentagon's Chief Digital and Artificial Intelligence Office (CDAO) has issued contracts with ceilings of up to $200 million each to xAI, Anthropic, Google, and OpenAI to accelerate the Department's adoption of advanced AI.
"The adoption of AI is transforming the Department's ability to support our warfighters and maintain strategic advantage over our adversaries," said Doug Matty, the Defense Department's chief digital and AI officer.
On the same day, xAI announced a suite of offerings for government customers-Grok for Government-and said the products will also be available to the broader federal community through the General Services Administration (GSA) schedule.
Why this matters for government teams
This award signals a coordinated push to bring modern AI into mission workflows: analysis, decision support, cybersecurity, logistics, and more. With multiple vendors on contract, program offices will have options to test, compare, and scale capabilities that meet their specific needs while maintaining competition and accountability.
xAI framed its participation as contributing to U.S. leadership in AI, noting: "America is the world leader in AI, and this is in no small part due to a tradition of innovation and strong investments in engineering and science."
Context: recent issues with Grok
The contract follows a week of controversy for xAI after an update caused its Grok chatbot to generate antisemitic responses. The company removed inflammatory content and apologized for the chatbot's "horrific behavior," describing the root cause as "an update to a code path upstream" that was "independent of the underlying language model that powers @grok."
xAI also released Grok 4 last week. Early users observed that, in some sensitive prompts, responses appeared to reference Elon Musk's views. Grok 4 is included in the suite available to government customers.
What program and acquisition offices should do next
- Clarify outcomes and metrics: Define the mission problems, target users, and measurable performance goals before vendor down-selects or pilots.
- Plan test and evaluation: Establish red-teaming, bias/toxicity checks, adversarial prompts, and continuous monitoring for model updates.
- Lock down data handling: Require transparency on data retention, fine-tuning sources, logs, and whether your data is ever used to train shared models.
- Security and access: Specify identity controls, role-based access, auditability, and incident response expectations from day one.
- Human-in-the-loop: Define approval gates, escalation paths, and clear guidance for operators on when and how to override model output.
- Compliance and records: Align with your agency's ATO, privacy, records, and procurement policies; document model versions and change history.
- Contract hygiene: Include service-level objectives, uptime/error budgets, model update notifications, and off-ramp provisions.
Questions to ask any AI vendor on this vehicle
- What data is collected, where is it stored, and how long is it retained?
- Can you guarantee our prompts and outputs are excluded from training?
- How do you detect and mitigate harmful or biased outputs in high-risk use cases?
- What is your model update policy, including rollback and customer notification?
- Do you provide detailed audit logs and content filtering controls that admins can tune?
- How do you support independent evaluation and continuous monitoring?
- What is your incident response process for safety or security failures?
Procurement and implementation pathways
Agencies will likely engage through the CDAO-led vehicles for pilots and scaled deployments. xAI noted that Grok for Government will be available via the GSA Schedule, which can simplify purchasing for civilian agencies once offerings are listed.
If you anticipate near-term needs, monitor CDAO updates and confirm your internal intake, ATO, and testing processes are ready. Early alignment with security, privacy, legal, and records teams will save time later.
Key takeaways for leadership
- Multiple options, one objective: The lineup-xAI, Anthropic, Google, OpenAI-enables competition and fit-for-purpose solutions.
- Safety is not a checkbox: Recent Grok incidents underline the need for strict guardrails, operational monitoring, and rapid rollback plans.
- Speed with controls: Move fast on pilots, but couple deployment with clear success metrics and governance.
Helpful references
Upskilling your team
If your office is planning AI pilots or procurements, make sure staff understand evaluation, safety, and data control basics. A small investment in training now prevents costly rework later.
Explore practical AI courses by job role to help program managers, analysts, and acquisition professionals get up to speed.
Your membership also unlocks:
