Engineers and proposal teams want AI for drafting and analysis. AILeakShield helps you find the AI in use, restrict what isn't approved, and stop marked CUI and credentials from reaching commercial AI tools.
Sample data shown. Select the image to enlarge it.
The risk sits in ordinary engineering and proposal work.
A spec or an extract from a drawing is uploaded for a quick summary.
Past performance or pricing is pasted in to tighten a bid.
Code is pasted into a chat tool to explain an error.
Internal addresses and system diagrams are shared while troubleshooting.
Part numbers, supplier pricing and program names appear in everyday prompts.
An AI assistant connected to Microsoft 365 can reach mail and files.
As of November 2026, DoD has suspended CMMC Phase 2, which would have required third-party assessments, and has announced no new date. Latham and Watkins summarizes what remains: Phase 1 self-assessments, FAR 52.204-21, DFARS 252.204-7012 and the NIST SP 800-171 controls for CUI all still apply. So the duty to protect CUI hasn't changed, and a commercial AI tool is outside your assessed boundary unless you've made it part of it. This is a moving area, so check the current DoD position.
AILeakShield isn't FedRAMP authorized, and it runs in commercial Microsoft Azure, not a government cloud. Text going to an AI tool is checked by our service first, so a prompt containing CUI reaches that service before it's blocked. DFARS 252.204-7012 requires FedRAMP Moderate equivalent security from an external cloud provider that stores, processes or transmits covered defense information. Treat AILeakShield as a control for general business systems, keep CUI work inside your enclave, and agree the approach with your assessor before you deploy to anyone who handles CUI.
One control on one channel, with evidence you can use.
Blocks CUI and other markings, credentials, personal identifiers and network details, and warns on financial and health data.
An inventory of AI sites, extensions, desktop apps and Microsoft 365 apps to inform your system security plan.
Show staff a clear notice on restricted tools, and let them request access.
Add program names, part numbers and codenames as sensitive terms.
Set stricter rules for engineering than for other teams, and turn on source code detection where it's needed.
Audit records you can map to the 800-171 access control, audit and system protection families, and send to your SIEM.
Start with steps that involve no prompt text.
The free Microsoft 365 report reads app connections, not email, files or chats.
Upload a DNS log to see which AI services are being contacted.
Decide with your assessor which users and systems are in scope.
Deploy to the agreed browsers, approve what you trust, and restrict the rest.
No. It addresses one channel. The other 800-171 controls, your system security plan and your assessment are outside this product.
No. It runs in commercial Azure. If you need a FedRAMP Moderate or equivalent service for covered defense information, AILeakShield isn't that service. Agree your approach with your assessor before deploying to anyone who handles CUI.
No. Marking detection needs the marking to be there. Unmarked CUI depends on the other detectors and your own terms, and an unmarked drawing can get through.
No. Text is checked by our service, so it reaches us before a decision is made. Blocked content is never stored, and other prompt text is cleared at the end of each day (UTC).
Yes. Manufacturers can protect formulas, designs and supplier pricing by adding their own sensitive terms and turning on source code detection. The CMMC pack is optional.
Coverage is managed browsers and the secure workspace. Files moved another way, unmanaged devices and personal phones are outside it.
Start with a free report of the AI connected to your Microsoft 365. It doesn't read any email, files or chats.