LumaLex Law · November 2026
If you run a business, you may have already received the call. A data broker, an AI training-data marketplace, or a lab’s sourcing team wants to buy your email archives, Slack history, support tickets, call recordings, code repositories, or customer records. The offers can be substantial, sometimes reaching six or seven figures, for data that’s simply sitting on your servers.
This trend started with startups that were shutting down and selling their Slack, email, and Jira archives to AI developers. Now, operating companies of every size are receiving similar offers. Gartner projects worldwide spending on AI data will reach roughly $6.5 billion in 2027, suggesting that demand will only continue to grow.
The problem is that the data you hold is rarely yours alone to sell. It may contain information belonging to your customers, employees, vendors, and, if you’re in healthcare, law, accounting, or another professional field, your patients and clients. The buyer’s term sheet is unlikely to address all of these concerns. Below are the legal issues we walk our clients through before they sign anything.
1. You May Not Have the Right to Sell It in the First Place
Start with your contracts, not your privacy policy. Your inbox and chat logs are full of other people’s confidential information shared under NDAs, MSAs, and partnership agreements. Customer contracts often limit how you can use their data. If you hold data as a processor or service provider for someone else, your ability to monetize it is generally restricted.
Platform terms matter, too. Slack’s API terms, for example, prohibit third-party apps from using API data to train large language models. Licensed content, vendor-generated records, third-party attachments, and open-source code can each carry restrictions or competing ownership claims.
Bottom line: Before discussing price, identify what data you hold, where it came from, and what each relevant agreement says about its use.
2. Your Privacy Policy Is a Promise, and the FTC Enforces It
Most privacy policies, employee handbooks, and customer notices never contemplated selling data to train AI. The general rule is that you must honor the privacy statement in effect when the data was collected.
The FTC has warned that quietly rewriting your policy to permit AI training or third-party sharing, then applying it to data you already have, can be an unfair or deceptive practice. In practice, this means previously collected data may require fresh, affirmative consent before it can be sold.
3. State Privacy Laws Treat This as a “Sale” With Real Obligations
As of late 2026, roughly two dozen states have comprehensive consumer privacy laws, and almost all of them regulate the “sale” of personal data broadly, often including any exchange for value, not just cash.
- California (CCPA/CPRA): Selling data that includes California residents’ personal information can constitute a sale under the law unless the data meets the statute’s strict de-identification standard. That standard requires technical safeguards, a public commitment not to re-identify the data, and contractual restrictions on the recipient. A sale also triggers applicable notice, opt-out, and risk-assessment obligations.
- Restrictions on sensitive data: States are moving beyond notice and consent toward outright prohibitions. Maryland and New Jersey prohibit the sale of sensitive data, and New Jersey’s ban applies regardless of consent and regardless of how many consumers’ data you process. Connecticut and Virginia now ban the sale of precise geolocation data, even with consent.
If your dataset includes information related to health, biometrics, precise location, children, sexual orientation, religion, or finances, it should be treated as particularly high-risk.
4. Selling Data Can Make You a “Data Broker”
Many businesses don’t realize that even a single data sale could bring them within the scope of state data broker laws. California, Vermont, Texas, Oregon, Connecticut, and New Jersey now require qualifying data brokers to register, although definitions vary by state.
California has particularly stringent requirements. Under the Delete Act, registered brokers must check the state’s centralized Delete Request and Opt-Out Platform (DROP) at least every 45 days and process deletion requests. Since August 1, 2026, a broker that fails to process DROP requests faces penalties of $200 per request, per day.
California brought its first enforcement action under both the CCPA and the Delete Act in August 2026, and the state’s broker registry now asks whether a broker has sold data to generative AI developers. New Jersey’s law carries penalties of $2,500 per day for registration failures and up to $50,000 per record for certain sensitive-data violations.
5. Healthcare: HIPAA and Beyond
For healthcare businesses, including medical practices, telehealth platforms, pharmacies, labs, and the vendors that serve them, the regulatory requirements are even more complex.
- Sale of PHI requires authorization. Under HIPAA, a covered entity or business associate that discloses protected health information in exchange for direct or indirect payment is generally considered to be selling PHI and typically needs a valid written authorization from each affected patient. Obtaining those authorizations is rarely practical at scale.
- Business associates face additional restrictions. If you’re a vendor holding PHI under a business associate agreement (BAA), your use of that data is limited to what the BAA and the Privacy Rule allow. Monetizing it for AI training is unlikely to be a permitted use.
- De-identification is the main path, but it’s more complicated than it sounds. Data properly de-identified under HIPAA’s Safe Harbor or Expert Determination method is no longer PHI. However, removing 18 identifiers from structured fields is very different from doing so reliably across millions of clinical notes, call transcripts, and chat messages. AI buyers also want linked, longitudinal records, which increases the risk of re-identification.
- State health privacy laws extend beyond HIPAA. Consumer health data laws, such as Washington’s My Health My Data Act, along with state sensitive-data bans, can reach health data that HIPAA doesn’t cover, including information held by wellness, fitness, and direct-to-consumer brands.
6. Law Firms, CPA Firms and Other Professionals: Confidentiality Comes First
Professional firms are increasingly being approached by data buyers because of the quality of their information. However, much of that information is also subject to extensive confidentiality requirements.
- Lawyers: ABA Model Rule 1.6 bars revealing any information relating to a representation without the client’s informed consent, subject to applicable exceptions, not just privileged material. Rule 1.9(c) extends that duty to former clients, and ABA Formal Opinion 480 confirms that information in public court filings can still be confidential under Rule 1.6. Realistically, a law firm considering a commercial data licensing arrangement would need informed consent from each affected client or a dataset containing no client information at all, such as certain internal templates and training materials. Selling client communications also risks waiving attorney-client privilege.
- CPAs and tax preparers: The AICPA’s Confidential Client Information Rule requires specific client consent, and the AICPA has cautioned firms to exercise care when evaluating AI training offers. Tax preparers face an additional layer of regulation under Internal Revenue Code Section 7216, which restricts the disclosure of tax return information and carries criminal penalties.
- Other licensed professionals, including financial advisors, insurance professionals, mental health providers, and engineers, should review their own regulatory and ethics rules, which often impose confidentiality duties stricter than general privacy law.
7. Employees and Labor Law
Workplace communications can contain highly sensitive employee information. Messages about pay, scheduling, and working conditions may constitute protected concerted activity under Section 7 of the National Labor Relations Act, whether or not your workforce is unionized.
Collective bargaining agreements may restrict a sale or trigger a duty to bargain. Removing employee names from a disciplinary discussion or pay complaint does not necessarily eliminate the legal or confidentiality concerns surrounding that information.
8. Trade Secrets, Privilege and Competitive Harm
Selling internal data can weaken trade secret protection for your pricing strategies, internal business documents, and source code. It can also risk waiving attorney-client privilege over legal communications contained in your email archives.
The buyer, or the buyer’s customers, could potentially use your data to build a competing product. Once data has been used to train an AI model, removing its influence from that model is generally difficult. Models can also reproduce portions of their training data in outputs to other users.
9. National Security Rules Apply to More Than You’d Think
The DOJ’s Bulk Sensitive Data Rule, now fully in force, prohibits U.S. companies from engaging in data brokerage(selling, licensing, or otherwise giving access to data) involving bulk U.S. sensitive personal data with six countries of concern, including China, or with covered persons tied to them.
It also requires contractual protections against onward transfer when brokering that data to any foreign person. Violations can carry civil and criminal penalties.
Separately, the FTC enforces the Protecting Americans’ Data from Foreign Adversaries Act, with civil penalties of up to $53,088 per violation, and has been sending warning letters to data brokers.
Understanding your buyer’s ownership, sources of capital, operational locations, and downstream customers is an important part of the due diligence process.
10. Your Sale May Not Stay Private
California’s AB 2013 and the EU AI Act require AI developers to disclose information about their training data sources, including whether data was purchased or licensed.
Your customers, employees, and competitors could eventually learn that you sold company data. The potential reputational consequences should be considered alongside the financial value of the transaction.
Before You Say Yes: A Practical Checklist
- Don’t send sample data until legal review is complete. Even sharing a sample constitutes a disclosure.
- Inventory the dataset: systems, record types, volume, date ranges, and the jurisdictions whose residents are represented in the data.
- Confirm your rights: review customer, vendor, partner, licensing, and platform agreements, along with any data you hold as a processor or business associate.
- Check your notices: What did your privacy policy, patient notices, engagement letters, or employee handbook say when the data was collected?
- Carve out high-risk data rather than relying on de-identification alone: privileged material, HR and payroll files, health and benefits data, client files, union communications, third-party confidential information, and regulated-system data.
- Control de-identification: jointly select the provider, approve the methodology, and require certification. For health data, consider HIPAA Expert Determination.
- Analyze registration and data broker status in every state where you have data subjects.
- Conduct due diligence on the buyer: ownership, foreign affiliations, downstream resale rights, and security practices.
- Document the decision: the legal review, exclusions, risk assessment, and business rationale. These records matter for regulators and future M&A due diligence.
Contract Terms Worth Fighting For
- Narrowly defined permitted uses (e.g., model training only, no resale, no re-identification, no competing products)
- Prohibitions on onward transfer, especially to foreign persons
- Robust buyer representations and warranties on security, compliance, and data use
- Indemnification and liability allocation that reflects who actually controls the data after closing
- Audit rights and certification of de-identification
- Clear procedures for what happens if a data subject, client, or regulator objects
- Confidentiality around the transaction itself, to the extent disclosure laws allow
There’s another consideration that businesses often overlook: your vendors may be receiving similar offers for your data. Review your SaaS, cloud, and service-provider agreements for clauses that allow them to sell or license “de-identified” or “aggregated” data. It’s also worth planning for what happens to your data if a vendor shuts down or goes bankrupt.
The Takeaway
Selling data for AI training isn’t automatically illegal, and for some businesses, it can be a valuable way to monetize an existing asset. However, most companies cannot sell everything a buyer asks for, and the compliance requirements are often more extensive than the initial offer suggests. Potential fines, professional ethics violations, and damage to customer trust can easily outweigh the financial benefits.
At LumaLex Law, we work at the intersection of emerging technology and heavily regulated industries, including healthcare, telehealth, life sciences, and professional services. If you’ve received an offer for your company’s data, or you want to understand your rights and obligations before one arrives, reach out to our team before sharing any files.
This article is for general informational purposes only and does not constitute legal advice. Reading it does not create an attorney-client relationship. Laws in this area are changing quickly; consult counsel about your specific situation.



