AI & ComputeLOT 261122 · 4 min read
Army plans OpenAI first in Project Gatekeeper AI phase
The Army plans to bring OpenAI into Project Gatekeeper next quarter, testing AI-assisted validation before a likely opening of the automated software accreditation platform.
Key takeaways
- Nov. 20: Chiulli said OpenAI will lead the first AI phase of Project Gatekeeper, targeted for the next quarter.
- Gatekeeper automates code checks for vendors facing authority-to-operate reviews that often take months.
- Passing code moves into an IL5 environment for controlled unclassified and mission-critical data.
- After the first phase, broader provider access is likely, with vendors supplying their own API keys.
- OpenAI said NATO uses up to 300 billion cyber-defense tokens monthly; no Army contract value was disclosed.

The U.S. Army plans to add AI models to Project Gatekeeper next quarter, with OpenAI in the first phase, acting Army CIO Gabe Chiulli said Friday, Nov. 20. Asked when AI integration would begin, Chiulli said: “I'm going to put it into the next quarter.” The statement sets a target, not a completed deployment.
Chiulli described a phased approach using different commercialized models, with OpenAI first. The Army disclosed no contract value, award or procurement vehicle for the project. OpenAI did not immediately answer DefenseScoop’s request for comment. Those omissions leave the commercial scale undefined.
What does Gatekeeper automate?
According to the Army, Gatekeeper, unveiled the previous month, lets commercial vendors submit code to an Army DevSecOps platform for automated checks and validation. Chiulli said software sold to the military needs an authority to operate, or ATO, a security accreditation process that often takes months before use on military networks.
The automated workflow is intended to reduce that delay, Chiulli said. Under his description, code that passes moves into an IL5 environment, the Defense Department’s impact level for controlled unclassified and mission-critical data. In Chiulli’s account, the government check lets a vendor sell or distribute its product into a secure Army enclave.
Neither Chiulli nor the Army supplied a measured time saving or vendor adoption figure. Chiulli said vendors holding a Defense Department Common Access Card can log in today. Officials are working to open access through Login.gov.
Chiulli also said he is working with other military departments’ CIOs on reciprocity, so each service does not build its own version. No timetable accompanied either access effort.
Why is OpenAI first?
Chiulli said the first phase was always intended for OpenAI because Gatekeeper sits on the commercial internet and uses services such as Login.gov. The Army did not describe a competitive selection process, evaluation criteria or whether another model provider was considered for the project in its first phase. Chiulli’s explanation points to architecture, not a disclosed bid.
OpenAI launched Daybreak Red Team and Defense Factory earlier that month. The company describes them as agents that penetration-test web applications and draft security patches. Chiulli said Gatekeeper validation will draw on tools such as OpenAI’s Daybreak. The Army named no specific Daybreak capability for the first phase.
Chiulli said: “The first phase is working with OpenAI to kind of work through, like, what do we want to check first?” The Army does not want to duplicate existing validation tools or consume tokens without a clear purpose. No benchmark, test-coverage target or expected error rate accompanied the announcement.
What changes after the OpenAI phase?
After the OpenAI phase, the Army will likely open Gatekeeper to other open-source and commercial model providers, Chiulli said. He described that expansion as likely, not committed. No start date, provider list or minimum technical standard accompanied that plan.
Chiulli said vendors would be encouraged to bring their own API keys and that each vendor would pay for the security of its own model. Under his account, each vendor keeps its code scan logs visible only to itself. That narrows Army visibility into code scans and shifts security costs to vendors.
Chiulli also said the Army plans calls for different harnesses to perform separate checks. No harness specification or provider timetable accompanied the proposal.
What is the commercial stakes?
The Army has published no vendor count, projected savings or completion schedule. That leaves the project without a measurable service commitment, despite Chiulli’s claim that automation should reduce months-long ATO reviews for commercial software on Army networks. The Army also has not said how many vendors have used the platform.
OpenAI separately said this month that NATO uses up to 300 billion tokens per month of its cyber-defense models. That company-reported figure measures NATO usage, not Army adoption. It does not establish Army demand, spending or the scale of the Gatekeeper project.
Chiulli’s next-quarter target is the next verifiable milestone. The Army must define its first validation checks before it can judge broader provider competition. Measured approval times, log visibility and vendor-funded API security will determine whether other model providers join after OpenAI.
mlq.ai (Source)
More from Sophie Lindqvist
Show full bio
News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,238 articles