Manufacturing
CAM and ERP software terms: can the vendor train AI on your programs?
By SourceX Editorial · Reviewed by Noah Loul ·
Short answer
Whether a CAM or ERP vendor can train AI on your programs depends on a few clauses read together: how the agreement defines customer data, the license you grant to run the service, and any usage data, aggregated data or product improvement terms. Check AI feature terms and admin settings separately, because they often carry their own rules.
Key takeaways
- Owning your files does not settle what the vendor may do with them; the license and data-use clauses do.
- Usage, telemetry and aggregated data clauses often sit outside the customer data definition and any promise attached to it.
- An explicit no-training clause in the contract is stronger than a statement on a vendor's blog or help page.
- Vendor terms can also restrict your own use of outputs or exports for AI, which matters before you license records.
- Customer models and defense work need their own decision, because customer NDAs and export rules apply on top of vendor terms.
What a CAM or ERP vendor could learn from your shop#
A CAM or ERP vendor could learn a great deal from what your shop stores in its software: NC programs and toolpaths, post-processors, tool libraries with feeds and speeds, setup sheets, routings, standard times, quotes and job costs. In a cloud product that content sits on the vendor's servers; in a desktop product, the vendor may still receive usage data about how the software is used.
Those records describe how your shop machines parts profitably, which is why training rights matter. Many programs are also built on customer models, so the answer affects your customers' confidentiality as well as your own.
This is general information, not legal advice. Terms change and negotiated agreements differ, so read your signed documents with counsel.
Clauses to find, and what each can permit#
A handful of clauses decide whether a vendor can train AI on your programs, and they are rarely in one document. The order form usually incorporates online terms, AI feature terms, a privacy notice and a data processing addendum by reference.
| Clause | What it can permit | What to look for |
|---|---|---|
| Customer data or customer content definition | Sets which records any protective promise covers | Whether programs, tool libraries and uploaded models are included |
| License you grant the vendor | Hosting, processing and copying needed to run the service | Words such as improve, develop or create new products |
| Usage, telemetry or service data | Collecting how you use the software, often outside customer data | Whether parameters, file contents or screenshots count as usage data |
| Aggregated or de-identified data | Use of combined data across customers, often with few limits | How de-identification is defined and whether your content can be included |
| Product improvement or machine learning | Training or tuning models on your data or usage | Whether it reaches customer content or only usage data, and any opt-out |
| AI feature terms | Separate rules for assistants and generative features | Whether prompts, uploaded files and outputs are used for training |
| Feedback clause | Free use of suggestions you send | Whether files or sample programs sent to support count as feedback |
| Subprocessors and third-party AI providers | Sending data to outside model providers | Whether those providers may retain or train on it |
| Changes to terms | Updates that bind you after notice or continued use | How notice is given and whether you can object |
| Restrictions on your use | Limits on using outputs or exports to train AI | Whether licensing your own records could breach them |
Why ownership language does not settle training rights#
Ownership language does not settle training rights because owning a file and controlling its use are different things. Autodesk's General Terms, for example, include a section titled You Own Your Work, under which the customer keeps ownership of files, designs, models and data it creates and uploads. What any vendor may do with such content is governed by the license and data-use clauses elsewhere in its agreement.
The clearest protection is an explicit no-training commitment in the contract itself. After users objected to updated terms in June 2024, Adobe stated that it does not train generative AI on customer content and said it would add that statement to its Terms of Use as a legal obligation. Zoom added a similar sentence to its terms in August 2023 after a backlash over earlier changes.
Definitions decide how far any promise reaches. Oracle's NetSuite terms define Customer Data as content the customer provides that is stored in, or run on or through, the cloud service. A no-training promise limited to customer data may not cover usage data, so read the two definitions side by side.
The restriction can also run against you#
Vendor terms can restrict your own use of AI, not only the vendor's. AEC Magazine reported that a broad clause in Autodesk's terms and acceptable use policy, read literally, would stop users from training their own AI models on output from an Autodesk application, and that the language had been in Autodesk contracts since May 2018.
That matters if you plan to build an internal model on your CAM output or license exports of your own records to an AI developer. Before either, check whether the CAM or ERP terms limit how you use outputs or bulk exports, and record what you found for each system.
A clause checklist for your next renewal#
The best time to fix data-use terms is before renewal, when you can still ask for changes. Work through the list with the signed order form and every document it incorporates.
On unilateral changes, FTC staff warned in February 2024 that a company quietly and retroactively changing its terms to allow AI training on consumer data may be engaging in unfair or deceptive practices. That guidance concerns consumers, but it is a reminder to read re-acceptance prompts from business software vendors closely too.
- Collect the order form, master agreement, online terms, AI feature terms, privacy notice and data processing addendum.
- Search for train, machine learning, artificial intelligence, model, improve, aggregate, de-identified, usage data, telemetry and feedback.
- Map each clause to the records it reaches: programs, tool libraries, uploaded customer models, ERP transactions and usage data.
- Check admin settings for AI features, data sharing and product improvement programs, and switch off what you do not need.
- Ask the vendor for written confirmation of what is used for training, and request a no-training clause for customer content.
- Check how term changes are announced and whether continued use counts as acceptance.
- Record findings in a vendor register with document versions, dates and contacts.
Customer models and programs need a separate decision#
Customer models and the programs built from them need a separate decision because your customer, not your vendor, sets the confidentiality rules. Many NDAs and purchase order terms limit disclosure to third parties, and a cloud CAM vendor that stores or processes customer models may count as one.
Defense and aerospace work adds export-control rules on top. Shops often keep those jobs in on-premises software or a separate environment, and record which customers allow cloud processing so programmers know where each job may live.
Illustrative example: a shop reviews terms before moving to the cloud#
Illustrative: a fictional CNC shop running desktop CAM and an on-premises ERP plans to move both to cloud versions. The owner and an outside lawyer read the new terms side by side with the old ones before signing.
They find that the CAM agreement protects customer content but lets the vendor use usage data and aggregated data for product improvement, and that a new AI assistant has its own terms covering prompts and uploads. The cloud ERP agreement includes a feedback clause broad enough to cover sample files sent to support.
The shop turns off optional data sharing, gets written confirmation that uploaded models are not used for training, stops attaching programs to support tickets, and keeps its aerospace customers' work on a separate on-premises seat. Every finding goes into a vendor register, which later shortens a review of which records the shop could license.
How SourceX treats vendor terms#
SourceX reviews software vendor terms in the Rights step of the SourceX five-step transaction, alongside customer contracts and employee notices, because use and export restrictions in CAM, ERP and other platforms can shape what a company may license. The SourceX Evidence Packet records the resulting licensing rights and permitted use, so a buyer can see which systems the records came from and on what terms.
Frequently asked questions
If we own our data, can the vendor still use it?
Potentially, yes. Ownership clauses confirm you keep title, but the license you grant the vendor, plus usage data, aggregated data and product improvement clauses, can still allow defined uses. Read the ownership clause together with the data-use clauses and any AI feature terms before drawing a conclusion.
Is aggregated usage data a real concern for a machine shop?
It depends on what the vendor counts as usage data. Click counts and error logs reveal little. Parameters, feeds and speeds, cycle times or file contents could reveal how you machine parts. Ask the vendor exactly which fields are collected and whether any part or program content is included.
Can a small shop negotiate these terms?
Sometimes. Large vendors rarely rewrite click-through terms, but resellers and enterprise sales teams can often provide written confirmations, addenda for AI features or settings that disable data sharing. A documented answer from the vendor is better than nothing and belongs in your vendor register.
Do the same questions apply to on-premises ERP?
Fewer of them. On-premises software keeps transactions on your servers, but support access, remote diagnostics, telemetry and add-on cloud modules can still send data to the vendor. Check support agreements and connected services, especially AI features added through updates.
What if the vendor changes its terms after we sign?
Check the change clause. Some terms apply updates on notice or on continued use, while negotiated agreements may fix terms for the subscription period. Watch renewal notices and re-acceptance prompts, which is where changes to data-use and AI terms tend to appear.
Sources
- Autodesk's General Terms include a section titled 'You Own Your Work' saying the customer retains ownership of files, designs, models, data sets and similar material it creates and uploads to an Offering. Source
- AEC Magazine reported that a broad clause in Autodesk's terms and Acceptable Use Policy, read literally, would stop users from training their own AI models on output from an Autodesk application, and that the language had been in contracts since May 2018. Source
- On June 6, 2024, Adobe published a clarification after users objected to updated terms language many read as allowing Adobe to use their content to train AI. Source
- In June 2024 Adobe stated that it does not train generative AI on customer content and that it would add that statement to its Terms of Use as a legal obligation. Source
- On August 7, 2023, after backlash, Zoom added to its Terms of Service a sentence stating it will not use audio, video or chat Customer Content to train AI models without consent. Source
- Oracle's NetSuite Terms of Service define Customer Data as content the customer provides that is stored in, or run on or through, the Cloud Service. Source
- FTC staff warned on February 13, 2024 that surreptitious, retroactive terms changes to allow using consumers' data for AI training may be unfair or deceptive. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.