DataPacts

Seller track · Lesson 3 of 9 · 9:22

Who owns it: employees, customers, contracts

What you'll learn

  • Four layers of rights in your data
  • Who owns what your staff made
  • Why customer contracts can block a sale
  • Which records stay out

Free lesson

Enter your email to watch this lesson and the rest of the course.

Transcript

You hold the records. Does that make them yours to sell? Not quite. Owning the records doesn't settle the rights of the people in them. Customers, employees and contributors keep claims through contracts, privacy law and employment law.

First, what a rights audit actually checks.

A rights audit looks at four layers for every dataset. Copyright in the works. Database right and trade secrets. Platform and API terms. And the people named in the records. A dataset can be licensed only when every layer allows it.

For each dataset you ask four plain questions. Who created it? Who owns it? Which terms limit its use? And which people appear in it? The answer decides whether the dataset is included, excluded or fixed first.

Start with who made it. Documents and code written by employees in the course of their job usually belong to the employer. Works by contractors and founders are different. Those need a written transfer.

That default comes from statutes in the United States, the United Kingdom and Germany. But an agreement can reverse them, and no court decision applying them to an AI data license is on record. So read the actual employment terms and assignments, not just the rule.

Copyright isn't the only right in a dataset. In the EU, a maker who shows substantial investment in a database can prevent extraction of the whole or a substantial part of it. That right can be transferred or licensed, and it runs for fifteen years from the following first of January after completion.

Next, the software your data lives in. A platform's terms can limit training use or resale, and the limit can depend on the export route. The same data can be free to use through one route and barred through another. So the audit records the route with each terms version.

Then there are other people's secrets. Trade secrets work through duties of secrecy. So a company holding a customer's confidential records cannot license them as its own.

The audit itself runs in steps. List each dataset and its source system. Name who created it. Collect the ownership evidence. Read the platform terms, then your customer and vendor contracts. And record a result for each dataset: include, exclude, or fix first.

Why bother? Because buyers make you promise. One data program's published terms show the full chain: a warranty about prior grants, an indemnity for privacy and IP claims, and a refund right that outlives acceptance. If the manifest is materially inaccurate, payment can be clawed back even after acceptance.

What if the audit finds a gap? Most gaps close one of four ways. Get a written assignment. Ask the platform, customer or partner for permission. Carve out the fields the terms don't allow. Or remove the data of people who haven't agreed. A gap you can't close takes that dataset out of the license.

Now, your own staff.

Say a buyer wants staff emails, chats or meeting records. Before a single message leaves, you usually have to tell the people in them. In some countries you also have to ask them, or ask their works council first.

Notice, consent and the collective step are three different duties. Notice tells staff before it happens. Consent asks each person, and they can refuse. The collective step brings in the works council or union, who are informed, consulted or must agree.

And whose messages sit in that archive? Current staff, former staff, applicants, contractors, and the customers and suppliers they wrote to. Each group has its own notice position. A former employee's mailbox needs the same notice as a current one.

The scale gets big fast. One bankrupt US airline's court schedule lists a hundred million emails across eighty thousand email accounts.

In the EU, the GDPR sets an order. Article 13(3) requires you to inform staff of the new purpose before further processing. In California, licensing employees' personal information for money is a sale under the CCPA, so employees hold the right to opt out.

Where a works council or union exists, the collective step comes on top of individual notice. Its strength ranges from information to a right of consent. In the Netherlands, a works council holds a consent right over rules on processing staff personal data. And in one US bankruptcy case, unions were among those who objected to the data sale.

A defensible order looks like this. Map who is in the data, including former staff and applicants. Pick the legal basis for each country. Complete any works council step. Then notify, ask where needed, and keep a record.

Now, your customers.

Customer data is often the most useful data a company holds, and the most restricted. You may license it only where your customer contracts, your privacy notices and privacy law all allow it. Any one of them can block it.

Customer restrictions sit in four places. The negotiated contract and its order forms. The data processing terms. Your online terms and privacy notice. And statutes and professional rules, which no contract can change.

Three kinds of clause do most of the blocking. Confidentiality: delivering a dataset to an AI buyer is a disclosure, whatever the price or purpose. Data processing terms: if you hold data on your customers' behalf, it usually isn't yours to license. And no-training promises in customer terms.

One point about vendor AI terms. Licensing to an outside buyer is a further disclosure the same customers have not been asked about. Many vendors now state whether customer data trains AI, with a ban, opt-in, opt-out or terms that differ by plan. Those cover the vendor's own training.

Which rule applies? Notice tells customers before their data is used for a new purpose. An opt-out lets them refuse. Consent requires them to agree first. Which one applies depends on the law, the data and the promise you made at collection.

Two examples. Under the California CCPA, licensing personal information for money is a sale that consumers may opt out of. And a US Federal Trade Commission staff post says a surreptitious, retroactive change to terms may be unfair or deceptive.

So can you just update your terms? Only going forward. Re-papering customer terms works forward only, and customers who refuse or stay silent are left out of the license. So the licensed set can shrink.

Last, the records that stay out whatever you do.

Some data can't be licensed at all. Company data cannot be licensed for AI training when a law bars its sale or disclosure, when a contract with its source forbids relicensing, or when the rights already belong to someone else.

Sector laws name specific records. Protected health information can't be sold without an authorization that states the payment. Tax preparers can't license client return information without the taxpayer's specific written consent. And a recording made without the required consent can't be licensed, because using an unlawful interception is itself a violation.

One rule is absolute. A US rule bars licensing bulk sensitive personal data to buyers linked to six countries of concern, even when the data is anonymized. It has applied since April 2025.

What about data you licensed in? You can relicense it for AI training only if that vendor's license allows sublicensing for that use, and most data and content licenses do not. One system vendor's terms say a customer must not repackage, resell, or sublicense any leads or data accessed through the services.

So what do you do with a bar? Check the records one by one, after ownership is settled. Licensing only the cleared subset, or not licensing at all, are both valid outcomes.

So, the recap. Check four layers of rights before any buyer sees a sample. Notice your staff, and ask where the law says so. Read your customer contracts before you rely on a license. And accept that some records stay out. Next up is lesson A4, De-identification without destroying the value.

Read and watch the full lesson free. Enter your email to unlock every lesson of the academy.

Sources on DataPacts