The DPO of an engineering firm gets a short question from the board: we want to see which AI tools are in use and show staff a notice when they open a tool we have decided on, so which lawful basis are we doing that under? Her answer fits in two sentences, and then the work starts.
For a private organisation the basis is usually legitimate interest. For a public authority it is public task. Employee consent almost never. And the basis on its own says nothing yet about whether the measurement is lawful as configured; that is what the balancing test decides.
Why is consent not a good basis here?
Because consent under the GDPR must be freely given, and in an employment relationship it rarely is. Someone who depends on their employer cannot easily say no without wondering what it will cost them. European supervisory authorities have said so consistently.
There is a practical reason too: consent can be withdrawn, and a security measure that stops the moment one employee withdraws is not a measure but an offer.
Which basis does fit?
| Basis (Article 6(1) GDPR) | Fits this deployment? | Why |
|---|---|---|
| Consent (a) | Rarely | Not free in an employment relationship, and revocable |
| Performance of a contract (b) | No | The employment contract does not require this measurement |
| Legal obligation (c) | Only where a law prescribes the processing itself | The GDPR asks for appropriate security but does not prescribe this measurement |
| Public task (e) | Yes, for public authorities | Legitimate interest is not available to public authorities for their tasks |
| Legitimate interest (f) | Yes, for private organisations | Information security and preventing data breaches are recognised interests |
The GDPR's recitals name network and information security explicitly as an interest that can be legitimate; keeping customer data out of an unassessed tool sits close to that. But "can" is not "is", and that is what the three steps test.
Is a count without a user id personal data at all?
What arrives at the server from a measurement of AI use is a domain, a day and a device, with no user id, no path and no content. That record is not on its own traceable to a person. On the device the observation does start at someone's browser, and in a team of four "one person opens a transcription tool daily" can still point at somebody. How that one field separates measuring from tracking is in AI monitoring versus employee monitoring.
The sensible line: treat the design as processing, document the basis, and let the absence of an id and of content count in the balance.
Something similar holds for the notices. A notice in the browser on a tool the organisation has decided about is shown locally, based on the domain. What is counted afterwards is the outcome (shown, alternative opened, continued anyway), without an id.
The marking of sensitive data while typing does process text: in working memory, on servers in Europe, and discarded afterwards. That is the processing of personal data, usually customers', where the basis has to be recorded in any case.
How does the three-step balancing test work?
The EDPB has published guidelines on it; the core is three cumulative steps.
Step 1: is the interest legitimate? Lawful, specific enough to write down, and present rather than hypothetical. "Knowing which AI tools are used, so we can decide per tool and keep customer data out of an unassessed tool" meets that. "Visibility of what employees do" does not: that is not an interest but a means without a purpose.
Step 2: is the processing necessary? Could the purpose reasonably be met with less? A per-person measurement is not needed to know which tools are in use, a count per tool is enough; storing content is not needed to show a marking, processing in working memory is enough. Every field you cannot justify undermines necessity.
Step 3: does the interest outweigh the employee's? Here you look at what employees reasonably expect, how intrusive the processing is and which safeguards accompany it. No user id, no content, no per-person breakdown, a floor for small groups, explanation in advance, a right to object that is actually set up: each of those safeguards shifts the balance. A notice that lets someone continue deliberately weighs differently from a system that takes over the action.
All three must pass. If step 2 fails because there is an id in the data that serves no purpose, a good interest will not rescue it.
Why does the outcome differ per organisation and per deployment?
Because three things move: who you are, what you switch on, and why.
A municipality measures under public task and has no balancing test in the sense of point (f), but faces the same necessity and proportionality questions; what that assessment looks like for a public body is in a DPIA for workplace AI. A healthcare organisation where health data can surface in prompts also has to designate an Article 9 exception for what incidentally sits in typed text. An engineering firm with only project data does not have that layer.
What you switch on weighs just as much. Department insights off or on, notices only for not-allowed tools or also for tools without a decision, classification of the first prompt into a use-case category on or off: each setting changes what is processed, and so the balance. An assessment made for the default settings does not cover the extended ones.
The engineering firm therefore writes the assessment as three short paragraphs, with the settings it has chosen attached: count per tool, department insights off, notices on for not-allowed tools, markings on in the supported apps. The DPO checks the text, an external lawyer reads it once, and the works council receives it together with the design. When the works council formally has a say is in when must the works council be involved in AI.
BeeSensible's own DPIA documentation records that the customer chooses and documents the basis; the common routes are legitimate interest with a written assessment, and public task for public bodies. The usage streams carry no user id, the dashboard has no per-employee breakdown, and department insights are off by default. That makes steps 2 and 3 lighter; it does not make them unnecessary.
These are the figures the balance is about: counts and outcomes, with no people in them.
