Skip to content
Outrich Data

Employee messages, monitoring notices and AI data licensing

Updated 2026-11-16 · 7 min read · By the Outrich Data team

Workplace data is full of employee communications, and whether a company may license them for AI training is not settled law. Monitoring notices usually cover security and compliance, not licensing. Sensible practice is to limit scope to work channels, exclude direct messages, de-identify, notify staff and take advice. This is general information, not legal advice.

This page is general information, not legal advice. Whether your company can license a given dataset depends on your contracts, your privacy notices and the law where you and your customers are. Review it with your own counsel before signing anything.

Why employee messages are the unresolved issue

Most guidance on selling company data focuses on customers. Yet the records AI labs value most, such as chat threads, email, ticket notes, code review comments and call recordings, are written or spoken by employees. Every message has an author, and many contain personal remarks next to the work.

Two questions get mixed together. The first is ownership. In the United States, work product created in the scope of employment generally belongs to the employer, and most employment agreements say so. The second is privacy. An employee can have privacy interests in a message even when the employer owns the system it was sent on.

No statute or leading case squarely answers whether an employer may license employee communications to an AI lab for training. The position has to be reasoned from monitoring law, privacy law, contract and labor relations. This guide sets out the considerations in general terms. It is not legal advice, and the answer for any company depends on its facts and its jurisdictions.

What monitoring notices usually cover

Many employers tell staff, in a handbook or an acceptable use policy, that company systems are for business use and may be monitored. The stated purposes are usually security, legal compliance, investigation of misconduct and protection of company property. Some policies add that employees should have no expectation of privacy on company systems.

A notice written for those purposes does not obviously extend to licensing message content to a third party. Reading messages to investigate a security incident is one kind of use. Supplying years of conversation to an outside company for model training is another. A regulator, a court or the workforce itself may see the second as outside what staff were told.

The practical point is that an existing monitoring notice is evidence, not a permission slip. A title search reads the notices in force over the dataset's time span and reports what they said. Where licensing was not contemplated, the gap can be narrowed by the mitigations described below.

US state laws on electronic monitoring notice

The United States has no general federal law requiring employers to give notice of electronic monitoring. A small number of states do. New York Civil Rights Law section 52-c requires private employers that monitor employee telephone, email or internet use to give written notice on hiring, obtain an acknowledgment and post the notice. Connecticut and Delaware have statutes that likewise require employers to give employees prior notice of electronic monitoring.

These laws regulate monitoring, not licensing. They do not say whether monitored content may be supplied to third parties. Their relevance is indirect: they show what employees were entitled to be told, and a company that did not give a required notice has a separate problem to fix first.

Other state rules can also apply. Laws on recording calls differ in whether one party or all parties must consent. Comprehensive state privacy laws vary in how far they reach employee data, and California's applies to employees and job applicants. Biometric laws can cover voice recordings in some circumstances. This list is not complete, and state law changes often, so confirm the current position with counsel.

EU and UK: legitimate interests and works councils

In the European Union and the United Kingdom, employee messages are personal data under the GDPR and its UK equivalent. Processing them needs a lawful basis under Article 6. Consent is generally considered unreliable in employment, because the imbalance of power makes it hard to show that agreement was freely given.

The basis usually examined is legitimate interests. It requires a balancing test: the employer identifies its interest, shows the processing is necessary for it, and weighs it against the rights and reasonable expectations of the employees. Licensing communications for AI training is a new purpose, so purpose limitation and transparency duties also apply, and a data protection impact assessment is likely to be expected.

Collective rights sit on top. In several member states, works councils have information, consultation or co-determination rights over systems that monitor staff or process their data. Germany is the best-known example. Proceeding without the council where its involvement is required can make the processing unlawful under national law. Truly anonymized data falls outside the GDPR, but the standard is high and free-text messages are hard to anonymize.

Union and collective agreements

Where a workforce is unionized, a collective bargaining agreement may address surveillance, recording or the use of employee data. Some agreements restrict new uses of workplace technology without bargaining. Even where the agreement is silent, a decision to license employee communications could be a subject the union expects to be consulted on.

The review is simple to describe: read the agreement, identify any clause on monitoring, data or technology, and ask labor counsel whether a duty to bargain or consult arises. Employee relations matter as much as the legal position. A workforce that learns about a data deal from the press reacts differently from one that was told first.

Practical mitigations

None of the steps below settles the legal question. Together they shrink the amount of personal content in scope and bring the use closer to what employees could reasonably expect. Outrich Data applies scoping and de-identification of this kind to the workplace data it buys, and the choices are recorded in the title report.

  • Scope to work channels: include project, support and engineering channels; leave out social and off-topic spaces
  • Exclude direct messages, which carry the highest expectation of privacy and the most personal content
  • Exclude sensitive functions: human resources, legal, health, grievance and union-related material
  • De-identify: remove names, handles, contact details and identifying context, and replace authors with consistent placeholders
  • Set a time window: leave out periods before the relevant notice was in place
  • Notify: tell employees in plain language what will be licensed, in what form and for what use
  • Consider an objection route for employees who do not want their messages included
  • Consult the works council or union where one exists
  • Record the decision: the assessment, the scope chosen and the reasons

Risk by channel

Not all employee communications carry the same exposure. The table gives a general picture of how different sources are usually handled. It is a starting point for discussion with counsel, not a rule.

SourceWhy it is sensitiveTypical handling
Public work channelsMostly work content, some personal asidesIn scope after de-identification
Ticket and case notesWork product, often with customer detailsIn scope after scrubbing both employee and customer identifiers
Code review commentsWork product tied to named authorsIn scope with author names replaced
Work emailMixed work and personal content; external partiesNarrow scope, such as shared mailboxes, or exclude
Direct messagesHigh privacy expectation; personal contentExclude
Call recordingsVoice is identifying; recording consent rules applyTranscribe and de-identify, or exclude
HR, legal and health channelsSpecial categories and privilegeExclude

What to document

A company that proceeds should be able to show its reasoning later. Keep the versions of the monitoring and acceptable use notices in force across the dataset's period. Keep the assessment of lawful basis where EU or UK staff are involved. Keep the scope decision, the de-identification method and the test results, and the notice given to employees with its date.

A buyer's diligence will ask for most of these, and a data title report will reference them. The law in this area is still developing, so revisit the position before each new license and take advice from employment and privacy counsel in each jurisdiction where the employees work.

Common questions

Does the company own employee messages sent on work systems?
In the United States the employer generally owns work product created in the scope of employment, and usually the systems too. Ownership does not remove employees' privacy interests or notice rights, which are a separate question. Other countries differ. Ask counsel.
Is a handbook monitoring clause enough to license messages for AI training?
It is uncertain. Monitoring clauses are normally written for security and compliance, not for supplying content to third parties. Most careful sellers narrow the scope, de-identify and give a specific notice instead of relying on the clause alone.
Do we need employee consent?
It depends on the jurisdiction and the facts. In the EU and UK, consent from employees is often viewed as unreliable and other bases are examined. In the US there is no single rule. This is a question for employment and privacy counsel.
Does de-identification solve the problem?
It reduces it. Removing names and identifying details lowers privacy risk and may take data outside some laws. Free text is hard to de-identify fully, so scoping out direct messages and sensitive channels remains important.

Find out if your data qualifies.

Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.

esc