Can defense contractors use ChatGPT? CUI, CMMC, and FedRAMP in November 2026

The Morse teamPublished November 11, 202611 min read

Yes, defense contractors can use ChatGPT, but not for controlled unclassified information (CUI) or export-controlled data in a personal account. As of November 11, 2026, DFARS clause 252.204-7012 lets a contractor put covered defense information only in a cloud service that meets "security requirements equivalent to" the FedRAMP Moderate baseline, and OpenAI's FedRAMP listing covers ChatGPT Enterprise and the API Platform, not the Free or Plus app. Ordinary work that holds no contract information, like a textbook question or a public spec, is a matter of your company's AI policy.

CMMC Phase 2, which would have made third-party Level 2 certification a condition of award for many contracts, was due on November 10, 2026, but the Department of War suspended it on July 13, and the program stays in Phase 1. The safeguarding rules in DFARS 7012 never paused.

On November 11, 2026, we read the CMMC program rule (32 CFR Part 170), the DFARS CMMC acquisition rule, DFARS 252.204-7012, DoD's FedRAMP Moderate equivalency memo, NIST SP 800-171, and the ITAR's encryption carve-out, plus the government AI pages of OpenAI, Anthropic, Microsoft, Google, and the FedRAMP Marketplace. The DoD CIO's, DoD's business.defense.gov, and OpenAI's own pages blocked our requests that day, so for those we read and cite archived copies on the Wayback Machine, and we say so each time. Microsoft's guide to its government clouds names CUI and DFARS outright, so a contractor can tell from one page which tenant its work needs.

This isn't legal advice. Your contract, your facility security officer, and your compliance office decide what applies to you.

Can defense contractors use ChatGPT for CUI?

Not the public app. DFARS 252.204-7012 says that if a contractor "intends to use an external cloud service provider to store, process, or transmit any covered defense information," it must "require and ensure" that the provider meets security requirements equivalent to the FedRAMP Moderate baseline and follows the clause's rules for incident reporting, malware, media preservation, and forensic access (48 CFR 252.204-7012). A chatbot that reads your prompt on its servers is a cloud service that processes it.

DoD's memo of December 21, 2023 sets the bar for "equivalent": a cloud service must "achieve 100 percent compliance with the latest FedRAMP moderate security control baseline" through an assessment by a FedRAMP-recognized third party, and DoD's equivalency rules "do not allow for POA&M's" left open from that assessment. A service already FedRAMP Moderate Authorized in the Marketplace "can be leveraged without further assessment" (DoD CIO memo, archived copy).

NIST SP 800-171, the standard DFARS 7012 points to, reaches cloud tools as well. Requirement 3.1.20 tells a contractor to "verify and control/limit connections to and use of external systems," and its discussion says this "addresses the use of external systems for the processing, storage, or transmission of CUI, including accessing cloud services (e.g., infrastructure as a service, platform as a service, or software as a service)." It counts "personally owned systems" and devices as external systems too (NIST SP 800-171 Rev. 2). So your own phone needs your company's approval before it touches CUI.

CUI isn't only what DoD sends you

Covered defense information has two halves in the clause. One is information "marked or otherwise identified in the contract" and provided by DoD. The other is information "collected, developed, received, transmitted, used, or stored by or on behalf of the contractor in support of the performance of the contract," where it falls in a CUI Registry category (48 CFR 252.204-7012). The clause's definition of controlled technical information includes "engineering drawings," "specifications," "technical reports," and "computer software executable code and source code." A test report you wrote last week for the contract can be CDI even though nobody at DoD has seen it.

Federal contract information (FCI) sits one level below. The DoD CIO quotes the FAR's definition, "information, not intended for public release, that is provided by or generated for the Government under a contract," and CMMC Level 1 asks contractors to meet 15 basic safeguarding requirements for it (DoD CIO, archived copy). Contract documents and emails that aren't public belong in the tools your company approves, even when they aren't CUI.

Did CMMC Phase 2 start on November 10, 2026?

No. Guides written before July still give November 10, 2026 as the start of Phase 2, which is when the rule said it would begin. On July 13, 2026, the Department of War announced "the immediate suspension" of the Phase II requirements, "which were originally scheduled to come into effect on November 10, 2026," and said "All Phase I self-assessment requirements remain firmly in place" (DoD Office of Industrial Base Growth, archived copy).

The regulation's text hasn't caught up. The eCFR's current 32 CFR 170.3, read November 11, 2026, still says Phase 2 "begins one calendar year following the start date of Phase 1" and that DoD "intends to include the requirement for CMMC Status of Level 2 (C3PAO)" as a condition of award (32 CFR 170.3). The suspension came by memo, so a guide that quotes the rule alone will get the date wrong.

For now, the DoD CIO says it "will enforce cybersecurity compliance with NIST 800-171 Rev 2 through self-assessments and select government-led assessments," and that the pause "does not eliminate the requirement for companies to protect information in accordance with DFARS clause 252.204-7012" (DoD CIO, archived copy). Revision 2 is the one to know, even though NIST withdrew it on May 14, 2024 in favor of Revision 3 (NIST). For AI tools, nothing about the pause loosens the cloud rule above.

Is ChatGPT FedRAMP authorized?

The business products are; the personal app isn't the listed product. On April 27, 2026, OpenAI said it had "achieved FedRAMP 20x Moderate authorization for ChatGPT Enterprise and API Platform," and that the authorization applies "subject to each agency's policies and authorization decisions" (OpenAI, archived copy). The FedRAMP Marketplace lists "ChatGPT Enterprise and API Platform" as "FedRAMP Certified" since January 9, 2026, on the 20x Pilot path at Class C (Moderate), with 0 agency authorization letters listed as of November 11, 2026 (FedRAMP Marketplace).

The listing names ChatGPT Enterprise and the API, so a Free, Plus, or Pro account you pay for yourself isn't the service it covers. A company Enterprise workspace is still subject to your compliance office's approval before CUI goes in.

ChatGPT Gov is a different product, built for agencies. OpenAI's January 28, 2025 announcement says agencies "can deploy ChatGPT Gov in their own Microsoft Azure commercial cloud or Azure Government cloud," which lets them manage frameworks like "IL5, CJIS, ITAR, FedRAMP High" themselves (OpenAI, archived copy). Whatever the version, OpenAI's terms for individuals, effective January 1, 2026, say "You must comply with all applicable trade laws, including sanctions and export control laws" (OpenAI, archived copy).

Which AI tools can hold CUI?

Only deployments your company buys, configures, and puts in its assessment scope. Each cell below is the vendor's own wording or a close summary, read November 11, 2026:

ITAR technical data is a separate question, and each vendor answers it differently, as we set out in which AI tools offer an ITAR environment.

Anthropic's page also says FedRAMP and Impact Levels "are certifications for cloud services," while "AI models are software components, not cloud services." The same model can sit in an approved deployment and in a personal app that isn't, so the deployment is what your compliance office checks.

Can defense contractors use Claude?

On DoW contract work, not as things stand. Anthropic's government page says that in March 2026 the Department of War "designated Anthropic a supply chain risk," and that the designation, as in effect after the court rulings, "applies to the use of Claude as a direct part of DoW contracts and subcontracts." It adds: "For DoW contractors, it does not limit their use of Claude outside their covered DoW work" (Anthropic).

The litigation is still running. Anthropic says a federal court in the Northern District of California ruled on August 27, 2026 that one designation was unlawful, and that on September 25, 2026 the D.C. Circuit, "in a 2-1 decision, upheld a separate designation" (Anthropic). A company that does both commercial and DoW work may need to keep Claude out of the DoW side only, and its contracts office will say where that line runs.

What about ITAR technical data in ChatGPT?

Keep it in the environment your export compliance office approved. The ITAR's carve-out for sending or storing technical data applies only when the data is "secured using end-to-end encryption" with FIPS 140-2 modules or AES-128 strength (22 CFR 120.54), and a chatbot can only answer text it has decrypted. We go through the carve-out, deemed exports, and what engineers paste in is ChatGPT ITAR compliant.

What can defense contractors use public ChatGPT for?

Whatever your company's AI policy allows that holds no contract information. Public material is the safe side: a published military standard, a datasheet, a textbook problem, a question about how a NIST requirement is usually read, or a cover letter for a job. Before you paste, ask whether the text is marked, came from the contract, or was made for it. If the answer to any of those is yes, it goes in the approved tool.

On a company laptop, assume your employer can see what you type into any chatbot, whichever account you use. Data loss tools can flag text pasted into AI sites, as we lay out in can my employer see my ChatGPT. Employees of agencies, rather than contractors, have their own rules, covered in can government employees use ChatGPT.

How Morse keeps a contractor's personal questions off chatbot servers

A consumer chatbot keeps your questions on its servers, tied to your name and email, for as long as its policy allows. On Morse, your chat history stays in your browser instead. It's for the questions that are yours: an SF-86 question you'd rather not ask a coworker, an offer from another firm, a health worry, or a standard you're learning on your own time. CUI, CDI, FCI, and export-controlled data stay in the systems your contract authorizes. Morse makes no FedRAMP, CMMC, ITAR, or other compliance claim, so no Morse mode belongs in your assessment scope, and material your employer's policy keeps out of outside tools stays out of Morse too.

You choose a privacy mode for each chat:

• Private, the mode you land in, for a question about a job offer or a benefits form. A new chat opens in Private, where you talk to open-weight models such as DeepSeek, Kimi, or GLM, or to Grok. Morse has zero data retention agreements with its inference, search, and other providers, so the model's host keeps nothing of what you ask. For providers that confirm zero retention reply by reply, Morse checks that confirmation and rejects any reply that comes without it.
• Anonymous, when you want GPT, Claude, or Gemini for a general question. Before your message reaches the provider, our proxy strips your account, your email, and your IP address, and the provider receives it from Morse's account instead, so it answers without knowing it's you. It may hold the text for a limited time to check for abuse, under its own policy, so write the question the way you would on a public forum: no company name, no contract number, no clearance level.
• E2EE, for the SF-86 draft or the health question you'd show nobody. Choose an open-weight model such as GLM, Kimi, or DeepSeek. Before anything is sent, your browser verifies the secure enclave's hardware attestation and seals the message so that only that enclave can open it, and Morse's servers pass along ciphertext they can't read. Third-party providers run the enclaves, not Morse, and the protection holds as long as the enclave hardware and attestation work as designed. E2EE comes with Pro, at $19 a month.

In every mode, your chat history is stored only in your browser, encrypted, and our servers don't keep your messages or log what you write. It locks when you sign out, and unlocking it takes two keys, one of which never leaves your device. Morse doesn't train AI on your content, sell your data, or show ads (privacy policy), and because we don't store chats, we usually have very little to give authorities. Use your own phone or computer, and sign up with a personal email rather than your company address. Clearing your browser or deleting your account removes your chats for good, and Morse can't recover them, so keep a copy of anything you need.

On your own device, send a first message from the Morse home page without an account, or start free with no card needed. Choosing your model, and E2EE chats, come with Pro, at $19 a month.