3 ms·
This overlooks one of the biggest under-reported problems GDPR has created: a large percentage of all industrial sensor data is "personal data" under GDPR. Thes
by jandrewrogers 7y ago
This overlooks one of the biggest under-reported problems GDPR has created: a large percentage of all industrial sensor data is "personal data" under GDPR. These are systems and companies that nobody associates with collecting or using personal data, because their business isn't about people, but the regulations have defined the scope broadly enough that there is universal agreement among their legal experts that are liable for not treating this data as "personal" under GDPR.
This raises some difficult challenges that the average Internet business doesn't have to deal with:
- Compliance with GDPR requirements for personal data in many industrial settings is operationally impossible. These aren't Internet ad tech databases.
- Some industrial systems aren't the kinds of things you can trivially upgrade to make them compliant in any case. We are talking embedded systems with operational lifespans measured in decades. In many cases there are other strict regulatory compliance requirements around the design and modification of these systems.
- The workloads and data models for some high-scale sensor data models make it technically impossible, given the current state of computer science and hardware, to comply with some obligations under GDPR when handling "personal" data. And for a much larger set of systems, it would be economically implausible even though theoretically possible.
- Sensor data infrastructure software often lacks the basic functionality required to support compliance, as the functionality that the regulators assumed exists for other purposes has no purpose in this context and therefore has never been implemented. There is a disconnect between what is required of the software users and what the upstream vendors can or are willing to provide. These aren't software companies.
- For some specific industry sectors, compliance costs disproportionately fall on EU-based companies by virtue of the fact that their primary operations are in a European country, even though they sell into a global market. That's an economic own goal.
This has become a Sword of Damocles over some industrial companies because their legal teams have studied their exposure to GDPR, identified substantial compliance obligations, and realized that compliance is effectively impossible. It is pretty clear to me that the regulators were so focused on Internet advertising companies and similar that they were completely oblivious to the unintended consequences for unrelated industrial sectors.
I've been studying this problem for a few industrial sectors for a couple years now. You have companies scrambling to find technology that often doesn't exist and in some cases requires hardcore computer science R&D before it could exist. And this is a business opportunity for someone to add a tax to what these companies produce. But the worst part is that this extremely expensive compliance exercise does almost nothing for personal privacy because most of this data was being collected for boring industrial applications.
- jorams 7y ago> a large percentage of all industrial sensor data is "personal data" under GDPR Could you give some examples? I can't think of anything off the top of my head.
- jandrewrogers 7y agoMany industrial companies measure the operational environment at massive scales now using multimodal sensor networks and platforms thanks to plummeting sensor costs -- LIDAR, hyper-spectral imaging, video, RF/radar, audio, remote sensing, chemical and particulate sensors, et al. The exact mix and scope of coverage varies with industry and company. Sectors are diverse and include automative, utilities, agriculture, oil and gas, logistics, etc. The sensor data is primarily used to manage risk, increase efficiency, improve safety, adapt to changing conditions, respond to incidents, do preventative maintenance, and similar. Any sensor platform that can detect the existence of an entity in space and time and is measuring space where people exist, which is most of them, is collecting personal data. A sophisticated party can reconstruct the identity of detected entities in the sensor data in a straightforward way. Typically, the sensor coverage inherently collects data on a large number of people from which it is impossible to obtain consent, whoever happens to be within or wanders into the sensor range. The detectable people in these sensor data models are analytical by-catch. I've demonstrated this to many organizations using diverse exhaust from industrial sensor systems never designed for that purpose. Some of these data models are incredibly large and fast moving, petabytes per day. Many of them collect data in federated environments that are severely bandwidth-limited and energy restricted; while the data model is very rich, there aren't enough local resources to do anything outside the designed scope. You can neither push compliance operations to the data, since there isn't enough compute, nor can you backhaul it to someplace that does. The aggregate data models can exceed an exabyte, so you aren't indexing where people are (that would be incredibly expensive) and any attempt to brute-force search to identify people for compliance purposes would effectively be a denial-of-service attack on the system. tl;dr: the scale and scope of external environmental sensing platforms increasingly used by industrial companies inherently allows you to detect the locations of many people in space and time that are unrelated to the business operation. The necessary scale and operational architecture of these systems make GDPR compliance technically implausible.