Can military use ChatGPT? GenAI.mil, personal phones, and OPSEC
The Morse teamPublished 9 min read
Yes, the military can use ChatGPT, but for duty work only through ChatGPT Mil on the Department of War's GenAI.mil platform, which added it on August 31, 2026 and is cleared for controlled unclassified information (CUI) at Impact Level 5. A personal ChatGPT account on your own phone is fine for your own life. As of November 2026, though, every rule we read keeps nonpublic DoD information out of it, on duty or off, however well the app encrypts.
We read the primary documents on November 14, 2026: the War Department's GenAI.mil releases and DoD News coverage, OpenAI's GenAI.mil announcement, the Super Intelligence Office's site, the Army CIO's generative AI memo, the Marine Corps' AI guidance and MARADMIN 018/26, the DoD OPSEC manual, the DoD Inspector General's Signal report, and UCMJ Articles 2 and 92. Several .mil sites refused our direct requests, so we read those pages through a text-reader service and link the originals. The Army and the Marine Corps deserve credit here: both published their AI guidance in full, and both discourage commands from banning the tools outright.
This isn't legal advice. For your own situation, ask your unit's OPSEC coordinator, security manager, or legal office.
Can soldiers use ChatGPT for military work?
Yes, through GenAI.mil. The War Department launched it in December 2025 with Google's Gemini for Government, saying "all tools on GenAI.mil are certified for Controlled Unclassified Information (CUI) and Impact Level 5 (IL5)" (War Department). Access requires a common access card on the department's nonclassified network (DoD News).
On August 31, 2026, the department added Starshield AI's Grok for Government and OpenAI's ChatGPT Mil. The department's release says ChatGPT Mil "supports document-heavy unclassified work across the department, including planning, policy, logistics and administration" for "more than three million department personnel" (DoD News, September 1, 2026). When it announced the deal on February 9, 2026, OpenAI said data processed on GenAI.mil "remains isolated to the government environment" and isn't used to train its public or commercial models (OpenAI).
The name is changing. As of November 14, 2026, the department's Super Intelligence Office site says "GenAI.mil is rebranding: access ChatGPT Mil, Gemini Enterprise, and Grok on SI Tools," and describes SI Tools as running in "approved Impact Level 5 (IL-5) and higher environments" (Super Intelligence Office). The login link still points to genai.mil, and most news coverage still uses the old name.
Is GenAI.mil approved for everything?
No. The Marine Corps spells out the limit: GenAI.mil "is approved for processing Controlled Unclassified Information (CUI). It is not approved for handling Personally Identifiable Information (PII), Protected Health Information (PHI), or classified information" (MARADMIN 018/26, January 20, 2026). The same message authorizes it for Marines, civilians, and contractors "using government-furnished equipment" and says the platform keeps "persistent chat histories." It adds that systems like CamoGPT that meet the requirements for government systems "continue to be authorized for their intended and approved purpose," and that NIPRGPT was "in the process of sunsetting."
An evaluation draft with a Marine's name and identifiers, or a question about someone's medical profile, doesn't belong in GenAI.mil, even though it's the approved tool, and belongs in a personal ChatGPT account even less.
Can you use ChatGPT on your personal phone?
Yes, for your own life. Neither service memo we read bans personal use. Both make you answer for what you type. The Army CIO's memo of June 27, 2024 says users "are responsible for information inputted into publicly accessible Gen AI tools and are subject to existing legal, cybersecurity, information, operational security, and classification policies" (Army PPM CIO-024). The Marine Corps' guidance of December 4, 2024 says the same, applies to the "Total Force," and adds that users "have no expectations of privacy" for what they put into a GenAI system (NAVMC 5239.1).
The rules turn on nonpublic DoD information, a much wider category than classified. DoD defines DoD information to include anything "acquired by DoD employees as part of their official duties or because of their official status," and it's nonpublic until it's cleared for public release (DoD Inspector General, quoting DoDI 8582.01). By that definition, your unit's training calendar or who deploys when is nonpublic, even with no marking on it.
What does the Signal case mean for AI chats?
It shows that encryption doesn't make an outside service approved. On December 2, 2025, the DoD Inspector General concluded that Secretary Hegseth sent "sensitive, nonpublic, operational information" over Signal on his personal cell phone, and that this did not comply with DoD Instruction 8170.01, "which prohibits using a personal device for official business and using a nonapproved commercially available messaging application to send nonpublic DoD information" (DODIG-2026-021).
The IG report quotes DoDI 8170.01: "Do not use non-DoD-controlled electronic messaging services to process nonpublic DoD information, regardless of the service's perceived appearance of security" (DODIG-2026-021). DoDI 8170.01 governs messaging, not chatbots, but the Army and Marine memos point to the same information and OPSEC policies, and a privacy-focused AI is still a service DoD doesn't control. The report also notes that some of the Secretary's messages "had auto-deleted because of chat settings," so investigators relied in part on other people's accounts of the chat.
What counts as OPSEC in an AI prompt?
Anything on your unit's critical information list, and the small details that add up to it. The DoD OPSEC manual defines critical information as "information about DoD activities, intentions, capabilities, or limitations that an adversary seeks," and indicators as data that "adversaries can interpret and piece together" to reach it (DoDM 5205.02). It tells components to "deny adversaries the opportunity to take advantage of publicly available information, especially when aggregated," and to review content for "sensitivity in the aggregate."
One question about a PCS timeline gives little away. Months of questions in one account, tied to your name (your report date, a field exercise you're packing for, the gym hours on post, a spouse's worry about a deployment) add up to the aggregated record the manual warns about. Your unit keeps a critical information list that, in the manual's words, is "used by all personnel in the organization to identify unclassified information requiring application of OPSEC measures." Ask your OPSEC coordinator for it before you decide what's safe to type.
Can you get in trouble under the UCMJ for using ChatGPT?
You can if what you put in breaks an order or regulation. Article 92 covers anyone subject to the code who "violates or fails to obey any lawful general order or regulation" or "is derelict in the performance of his duties" (10 U.S.C. 892). The Army memo calls itself "clarifying guidance," and both memos say misuse of government software is handled under existing policy. Ask your legal office whether a given rule is punitive.
A clearance raises the stakes off duty too. An article in the Army's Gray Space journal of July 1, 2026 warns that "for a cleared professional, a personal compromise can surface as a security concern that affects a clearance, an assignment, and a unit's readiness" (Army, Line of Departure).
Do the rules apply to Guard and Reserve members off duty?
The information rules do, whatever your duty status. The Army CIO memo applies to "the Army National Guard/Army National Guard of the United States, and the Army Reserve," and the Marine guidance covers the Total Force. The UCMJ reaches you more narrowly: reserve component members are subject to it while on inactive-duty training and related travel, and Army and Air National Guard members "only when in Federal service" (10 U.S.C. 802). So the UCMJ may not reach a question you type on a Tuesday night between drills, but nonpublic information you learned at drill stays nonpublic all month.
Who can see a service member's AI chats, and what helps?
Each row names who might read what you type, from the sources above, read November 14, 2026.
| Who you're guarding against | What helps | What doesn't |
|---|---|---|
| Foreign intelligence piecing together indicators | Keeping unit details, dates, places, and names out of every chat; following your critical information list | Encryption or a VPN: DoD's messaging rule looks past a service's "perceived appearance of security" |
| The AI company behind a personal account | Turning off model improvement in ChatGPT's Data controls; temporary chats, which OpenAI "may retain" up to 30 days (OpenAI) | Clearing chats later: they're deleted within 30 days unless kept "for security or legal reasons" |
| Your command, on government equipment or GenAI.mil | Asking personal questions on your own device and account | Expecting privacy: users have "no expectations of privacy" for their inputs, and GenAI.mil keeps chat histories |
| An investigator or court | Less stored in the first place | Auto-delete alone: the IG rebuilt the Signal chat partly from other participants |
| An AI agent on your home network | Keeping it off the device or network your VPN client uses | Assuming it's harmless off post: "The agent is not on the Army network, but it is not entirely off of it, either" (Army, Line of Departure) |
How courts reach stored AI chats is in can AI chats be subpoenaed, and what a phone search at the border can open is in AI privacy at the border. Federal civilians at DoD follow agency rules covered in can government employees use ChatGPT, and contractors handling CUI have their own in can defense contractors use ChatGPT.
How Morse keeps a service member's chat history out of an AI account under their name
The aggregation problem starts with where a consumer chatbot keeps your chats: on the company's servers, in an account under your name, where OpenAI says authorized staff may read content to investigate abuse, handle legal matters, or improve its models unless you opt out (OpenAI). That stored file of your questions is what a court can order produced. Morse is built so your history isn't kept there.
Morse is for your own life, on your own device and your own time: a VA home loan, a budget before a PCS, a question about a divorce or a debt you'd rather not take to your chain of command. Classified information, CUI, anything on your unit's critical information list, and any other nonpublic DoD information stay out of Morse, as they stay out of every AI outside the tools your service approves. E2EE doesn't change that, for the reason the Signal report gives. Pick the mode for each chat:
- Private, so AI hosts have agreed to keep no file of your questions. Every new chat starts here, on open-weight models like DeepSeek, GLM, and Kimi. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Otherwise Morse relies on the provider's commitment.
- E2EE, for the personal question you'd show no one. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the protection rests on the enclave hardware and attestation working as designed. E2EE has no web search or image generation.
- Anonymous, to ask GPT, Claude, or Gemini without your name on the request. Our proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider answers without knowing it's you. It may keep the message for a limited time to check for abuse, under its own policy, so leave your unit, rank, and installation out of what you write.
In every mode, your chat history is stored in your browser, encrypted, not on Morse's servers, and it locks when you sign out. Our servers don't keep your messages or log what you write. 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. Clearing your browser or losing the device deletes your chats, 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. E2EE chats come with Pro, at $19 a month.
Common questions
Can I use ChatGPT to write my evaluation or award bullets?
Only in a tool your command allows, and only with what that tool is approved for. An evaluation carries names and other personal details, and the Marine Corps says GenAI.mil itself is not approved for PII, so ask your rater or your command's AI lead before you paste a draft anywhere.
Can I log in to GenAI.mil from my personal phone?
Not as the services describe it. DoD News says access requires a common access card on the department's nonclassified network, and the Marine Corps' MARADMIN 018/26 authorizes GenAI.mil for Marines, civilians, and contractors "using government-furnished equipment."
Is CamoGPT still allowed now that GenAI.mil exists?
For the Marine Corps, yes. MARADMIN 018/26 says systems like CamoGPT that meet the requirements for government systems "continue to be authorized for their intended and approved purpose," and it said NIPRGPT was "in the process of sunsetting."
Does OpenAI train on what I type into ChatGPT Mil?
OpenAI says no. Its February 9, 2026 announcement says data processed on GenAI.mil "is not used to train or improve OpenAI's public or commercial models." That promise covers the GenAI.mil version, not a personal ChatGPT account.