Is Radio Monitoring in a Workplace Legal?
· 6 min read · Updated
Radio monitoring in a workplace means processing personal data, because a device's hardware address, the MAC address that every Bluetooth or Wi-Fi chip carries, has been treated as personal data by European supervisory authorities for over a decade, and what follows from that differs more from one country to the next than the marketing in this field lets on, so we say up front that we are engineers and not lawyers, and that what follows is how we understand the rules we design against rather than legal advice.
Last of four articles on radio detection in industrial environments, after what you can hear, what you can find and how to tell when you are being deceived, and this one covers whether you are allowed to do any of it.
Why this article exists
Almost every technical article on radio detection skips this chapter, and we think that is a mistake with a predictable ending, because a system deployed on sound engineering and then discovered by a workforce that was never told tends to be removed within a quarter. We prefer to treat the question as an engineering constraint rather than a legal afterthought, because the rules shape the architecture and not only the paperwork.
The European position and the CNIL guidance
A device's hardware address is personal data in the eyes of European supervisory authorities, who have been consistent on this point since well before the GDPR, and the reasoning is simple enough, because an identifier that persists from one observation to the next and attaches to a device carried by one person singles out that person whether or not you know their name. The most detailed published guidance we know of in Europe comes from the French supervisory authority, the CNIL, and it is worth knowing wherever you operate because it remains, as far as we can tell, the most specific document on this exact use case, and in substance it asks the following. For simple aggregate counting it asks for anonymisation within five minutes of collection, and the word is anonymisation rather than pseudonymisation, which means the operation has to be irreversible, or alternatively for immediate pseudonymisation followed by anonymisation or destruction within twenty-four hours, together with a markedly fuller notice to the people concerned. It asks for visible signage on site, on the model of CCTV notices, pointing to fuller information, and it asks for a workable right to object, and on this last point one clarification settles much debate, because asking people to switch off their phone's Wi-Fi was not considered an acceptable way to object, since it deprives them of a function of their own device. The history is instructive too, because in 2015 a large outdoor advertising company was refused authorisation for a Wi-Fi counting system in a Paris business district on the grounds that the anonymisation it proposed was reversible, as the operator could replay the hashing process, and reversible anonymisation is pseudonymisation under a better name, and the regulator was not taken in by it.
Where the picture differs
Two caveats are worth stating, because the field is not uniform and vendors who sell globally tend to imply that it is. Outside the EU and the UK the position is far less settled as far as we know, because there is no US federal equivalent to the GDPR covering this case, some state privacy laws include device identifiers in their definition of personal information while others do not, and sector rules apply where they apply and stay silent elsewhere. That looks like a compliance advantage today and a liability tomorrow, because the privacy regimes we have watched have tightened rather than loosened, so a system built to exploit a permissive jurisdiction will probably need rebuilding when that jurisdiction catches up. The workforce dimension varies substantially as well, because in several European jurisdictions deploying radio monitoring in a workplace triggers a formal employee consultation obligation before installation and not after, while in others there is no such requirement, and our recommendation is the same either way, which is to treat it as an obligation, because a system discovered by the workforce rather than presented to it rarely survives its first meeting with staff representatives, and the engineering merits of the architecture will not be part of that conversation.
What we do about radio monitoring in a workplace
Our position follows from the architecture more than from virtue, and we think that is the only kind of privacy claim worth making. We do not need to track strangers, because the system answers one question, which is whether the enrolled holder is present, so the identifiers we retain are those of voluntarily enrolled badges and phones whose holders are informed and whose pairing key was handed to us explicitly, and for everything else, including the anomaly detection described in the third article, an aggregate counter and a signature comparison are enough. The rotating address mechanism, which phone makers call MAC randomization, works in our favour, because as we explained in Who Is in the Room? an unknown Bluetooth address cannot be linked from one cycle to the next without the pairing key, so not tracking third parties is a property of the protocol rather than a contractual promise, and when someone asks whether our system tracks visitors' phones the answer is that it cannot, which we find better than a promise that it does not. Our logs name authorised people and not passers-by, because an anomaly event records that an unidentified emitter was present, with its timestamp and its band, and it does not reconstruct a third party's identity since it has no means to. The same rule applies to our keyboard and mouse effector, which logs the event, whether a connection, a cut or an unauthorised pairing attempt, and never the content, because recording keystrokes builds a keylogger however good the intention, and we doubt any security justification would survive that description in front of a data protection officer.
Three questions to ask any vendor in this field
They are useful whoever you are evaluating, including us. The first is what identifiers you retain and for how long, because if the answer involves storing raw hardware addresses of unenrolled devices for any period beyond immediate processing you should ask what the lawful basis is and expect a precise answer. The second is whether the system can track a device you have not enrolled, because if it can that is a capability that has to be justified and governed and disclosed, while if it cannot as a matter of protocol rather than policy that is worth considerably more. The third is what appears in the log when an unidentified device is detected, because a timestamp, a band and a signature class is the kind of answer we would expect, while anything from which a person could be re-identified is a wrong one. If a vendor cannot answer these three without checking with someone, the system was probably designed before the question was asked.
The engineering point
The thread running through this series is that radio detection corroborates a decision rather than anchoring one, and the data protection position follows from exactly the same place, because a system that needs to identify everyone in a room in order to work has taken on a legal burden it did not need, while a system that identifies only those who volunteered to be identified and treats everything else as an unattributed anomaly signal does the same operational job with a fraction of the exposure. That is not a compromise we made under regulatory pressure, since it is the architecture that fell out of asking what the system needed to know, which turned out to be much less than what it could know.
