9
If your firm handles controlled unclassified information (CUI), public AI is a hard stop — not a preference, a compliance boundary. CMMC 2.0 and NIST SP 800-171 require you to protect CUI with specific controls. And CUI cannot be transmitted to systems that aren’t approved for it.
What CUI means in practice
CUI is the government’s label for information that, while not classified, still requires safeguarding — contract data, technical drawings, controlled technical information, and the records that flow through a government contractor’s daily work. The rules that govern it are not suggestions. They are contractual and regulatory obligations with real consequences for noncompliance.
NIST SP 800-171 lays out the controls: access limits, audit trails, transmission protections, and the requirement that CUI reside only in authorized systems. CMMC 2.0 turns those controls into a certification framework that primes, subcontractors, and primes’ supply chains must meet to win and keep work.
What public AI does to CUI
Paste a paragraph of CUI into a public AI tool, and you’ve just moved controlled data onto someone else’s unapproved infrastructure. That’s a compliance failure — and in the government-contracting world, that can cost you a contract, a certification, or worse.
The tool is not an approved system. It has no NIST SP 800-171 controls. It offers no audit trail you can produce in a certification assessment. The moment CUI touches it, the data has left the boundary the rules were written to protect — and the firm is left to explain how controlled information ended up on a third party’s servers.
The certification stakes
This is not an abstract compliance concern. CMMC 2.0 certification is increasingly a condition of doing business with the Department of Defense and other agencies. A firm that cannot demonstrate it protects CUI — including how it uses AI — may find itself locked out of the very contracts it depends on. The assessment looks at your systems, your controls, and your data flows. A public AI tool that has touched CUI is a finding waiting to be discovered.
The cost of getting this wrong compounds. A failed assessment delays certification. A compliance failure discovered after the fact can trigger a corrective-action plan, a loss of eligibility, or a breach notification that reaches the contracting officer. None of these are hypotheticals. They are the ordinary consequences of moving controlled data onto unapproved infrastructure.
Why “just use ChatGPT” is wrong here
This is why “just use ChatGPT” is categorically wrong for this industry. The capability you want — drafting, analysis, proposal writing from your own history — is real. But it has to run on infrastructure you control, where CUI never leaves your boundary.
Open-weights models give you that. You run them on your own systems, behind your own controls, with a data trail you can audit. Same capability, compliant by design. The model is a tool inside your approved environment, not a channel that carries controlled data out of it.
The question for every contractor
If you’re a government contractor, the question isn’t whether you can use AI. It’s whether you can use it without breaking CMMC. The Confidential AI Exposure Checklist shows you where the risk is. Nine lives. Zero leaks.