Decision tree · cleared use

Cleared AI Use Decision Tree

Run this six-step check before you paste anything into a public AI tool. If any step lands on YES, stop. The whole tree takes about ten seconds once you know it. Print it. Tape it next to your monitor.

The six steps

Step 1. Is the content classified?

  • If YES: Stop. Reason: Classified material does not go into commercial AI tools. Ever. This includes Confidential, Secret, Top Secret, and SCI. Mishandling is a felony and a clearance-ender.
  • If NO: continue to step 2.

Step 2. Is it CUI?

  • If YES: Stop. Reason: Controlled Unclassified Information has handling rules. Most commercial AI tools do not meet them. Use only an approved on-prem or GovCloud-class tool with a documented agreement.
  • If NO: continue to step 3.

Step 3. Is it ITAR or EAR controlled?

  • If YES: Stop. Reason: Export-controlled technical data sent to a tool that may route to foreign infrastructure is a deemed export. The tool's server location and staff nationality matter. Assume commercial tools fail this test.
  • If NO: continue to step 4.

Step 4. Does it contain another person's PII?

  • If YES: Stop. Reason: Names, SSNs, DOBs, addresses, medical or financial details about someone else. Privacy rules apply even when nothing is classified. Strip it or do not paste it.
  • If NO: continue to step 5.

Step 5. Does it reveal a program, system, or capability that is not publicly disclosed?

  • If YES: Stop. Reason: This is the gray-zone trap. The sentence may contain no marked information but still expose a program name, a contractor pairing, a fielding timeline, a sensor capability, or a known-but-not-public location. Aggregation is real. Adversaries collect this.
  • If NO: continue to step 6.

Step 6. Is it company-confidential or covered by an NDA?

  • If YES: Stop. Reason: Pricing, proposals, customer lists, source code, internal strategy. Your employer's rules apply on top of government rules. Use only the tool your company has approved for that data class.
  • If NO: Proceed.

Why step 5 matters

Most cleared incidents do not happen at step 1. People know not to paste a SECRET document into ChatGPT.

Incidents happen at step 5. Someone pastes a paragraph that, on its face, contains nothing marked. But the paragraph names a program, a vendor, a deployment site, or a capability. The information is technically unclassified. The aggregation is not.

Step 5 is the test that separates engineers who keep their clearance from engineers who lose it.

If you are not sure whether something is publicly disclosed, the answer is no. Ask your security officer. Treat "I think it's public" the same as "I do not know."

What "Stop" means

Stop means:

  1. Do not paste.
  2. Do not paraphrase and paste. Paraphrasing classified or controlled information does not declassify it.
  3. Use an approved on-prem or air-gapped tool, or do the work without AI.
  4. If you already pasted, see the incident self-report reference.