For owners & operations leaders
Mid-market manufacturing · 100–1,000 employees
The AI Security Checklist for Manufacturers
Your people are already using AI. The question is whether your quotes, your pricing, and your customers' drawings stay in your building — or end up somewhere you've never heard of, under terms you never read. Ten minutes with this checklist tells you which.
Voice of the Customer
What owners and CTOs tell us when AI comes up. Paraphrased from real conversations.
Our quotes, our drawings, our process knowledge. That's one of the most valuable things this company owns. Nobody should be able to train on it for free, and we don't trust the big AI companies with it. We don't want our data showing up in a chat one of our competitors is having with a public AI.
President & Owner — industrial equipment manufacturer
If we ever part ways with a vendor, the data and the code stay in our building. That's not a preference. That's the term the deal lives or dies on.
CTO — industrial machinery manufacturer
01
Find out what's already in use
You can't secure what you can't see. Before you write a single policy, get an honest inventory — and make it safe to be honest. Punish people for admitting it, and the next survey comes back clean while nothing changes.
Survey every department on which AI tools they use and for what — quoting emails, drawing review, spreadsheet cleanup, customer correspondence.Anonymous works better than attributed. You want truth, not compliance theater.
Check your network and expense reports for AI subscriptions you didn't authorize.A $20/month ChatGPT Plus charge on a company card is a data pipeline you didn't know you had.
Ask specifically about free-tier accounts. Free versions of AI tools typically have weaker data terms than paid enterprise plans.Many free tiers reserve the right to train on what users type in. Enterprise tiers usually don't.
02
Classify your data before any tool touches it
Your most valuable data isn't in a vault — it's in quote folders, travelers, and process knowledge that makes your yield better than your competitor's. Decide what's restricted before someone pastes it into a chatbot.
Define three tiers: public (your website), internal (pricing, margins, quotes, customer lists), and restricted (export-controlled technical data, anything under NDA).
Mark customer drawings and specs as restricted by default.Most of it isn't yours to share. Your customer's NDA almost certainly counts an AI vendor as a third party.
Write one plain-language rule everyone can recite: "If you wouldn't email it to a stranger, don't paste it into an AI tool we haven't approved."
03
Put export control in its own category
If you touch defense, aerospace, or anything dual-use, this section outranks the rest. Export violations carry criminal penalties, and "an employee did it on his own" is not a defense.
Prohibit ITAR and EAR-controlled technical data in any consumer AI tool. No exceptions.Uploading controlled data to a system accessible by foreign persons can constitute an unauthorized export — even if no one ever reads it.
If you hold DoD contracts, map AI tools against NIST SP 800-171 and your CMMC obligations.CMMC requirements are now phasing into defense contracts. An unvetted AI tool handling CUI can put your assessment — and your contract eligibility — at risk.
Review your customer flow-down clauses. Primes pass their security obligations down to you, and AI usage is increasingly named explicitly.
The uncomfortable truth: a junior estimator pasting a controlled drawing into a free chatbot to "summarize the requirements" may have just committed an export violation on your behalf. Policy is not optional here.
04
Vet vendors before the demo wins you over
Every AI vendor will tell you your data is safe. The ones worth working with will put it in writing and let you verify it. Ask these questions before pricing — the answers tell you who you're dealing with.
"Do you train models on our data?" Get the answer in the contract, not the sales call."We may use your data to improve our services" is a yes wearing a suit.
"What happens to our data when we leave?" Deletion timelines, in writing, including backups.
05
Control the doors
Most AI data leaks aren't sophisticated attacks. They're an open account, a shared password, or a former employee who still has access. The fixes are boring and they work.
Require single sign-on and multi-factor authentication on every approved AI tool. No personal accounts for company work.
Turn off model training in every AI tool's settings. ChatGPT, Claude, Gemini, and Copilot each have a data or privacy switch for it.Their policies say they don't train on business accounts. That's hard to verify from the outside, so flip the switch yourself and keep restricted data out regardless.
Set role-based access. The person quoting doesn't need engineering's full drawing library inside the AI tool, and vice versa.
Add AI accounts to your offboarding checklist.Departing employees keep access to forgotten tools more often than anyone admits — along with everything they ever uploaded.
06
Get it in the contract
The terms of service you clicked through are written for the vendor. The contract you negotiate is written for you. For any tool touching internal or restricted data, negotiate.
Sign a data processing agreement (DPA) that names what the vendor can and cannot do with your data.
Confirm you own the outputs — the quotes, summaries, and documents the tool generates from your data.
07
Write a policy people will actually follow
Banning AI doesn't stop AI use — it just pushes it underground, where you can't see it. The companies that handle this well give people a sanctioned path that's better than the workaround.
Publish an approved-tools list and keep it current. If the answer is always no, people stop asking.
Keep the policy to one page. What's approved, what's restricted, who to ask. Nobody reads a 30-page AI governance document on a production floor.
Create a no-blame reporting path. Someone who pasted the wrong thing should tell you within the hour, not hide it for a year.
How to read your score
16+ checkedYou're ahead of most of your industry. Keep the review on the calendar.
10–15 checkedSolid foundation, real gaps. Start with Sections 03 and 04 — compliance and vendor terms carry the biggest downside.
Under 10Your data is making decisions without you. Block an afternoon this month. Section 01 first: you can't fix what you haven't found.
Why Annora cares about this
Your knowledge is the asset. Treat it like one.
Annora helps mid-market manufacturers capture institutional knowledge — quoting logic, process specs, the answers that live in your senior people's heads — and put it to work without pouring it into tools you don't control. Built for manufacturers, with data terms written for the questions in Section 04.
Book a 20-minute review call and we'll go through this checklist against your operation: where you're exposed, what to fix first, and what has worked elsewhere. Nothing to sign.
Book a review call →