If you’ve used ChatGPT, you know it types an answer back. You ask a question, it writes an email, or maybe it summarizes a document. That’s a powerful trick.
But a newer kind of AI doesn’t just talk. It acts. This isn’t just about generating text; it’s about AI taking steps, making decisions, and using other tools to complete a task. Think of it as a digital assistant that doesn’t need you to tell it every single click or search. It figures things out on its own.
Here’s the thing: this new capability is quietly changing what we thought was private, especially about where people go.
The silent threat to your most personal data
For years, companies have collected vast amounts of location data. This is often called mobility microdata (just a fancy way of saying very detailed information about where you move, often down to GPS coordinates and timestamps). Think of all the apps on your phone that ask for your location, or the smart devices in your car. This data is incredibly valuable for everything from traffic planning to targeted advertising.
Most companies promise to make this data “anonymous.” They scrub out names, email addresses, and phone numbers. They believe that if they remove direct identifiers, the data is safe and can’t be traced back to any single person. This idea of de facto anonymity (meaning, it’s anonymous enough in practice) has been a cornerstone of data privacy practices for a long time.
But that’s where the wrong answer comes in. Most people, even many data professionals, assume “anonymous” means truly untraceable. At first glance, it makes sense. If my name isn’t there, how could anyone know it’s me? This assumption has been a comforting, if ultimately flawed, foundation for how we share and handle personal movement patterns.
How a digital detective can find you
The real answer is that your movements are incredibly unique. Even if your name is removed, the specific pattern of places you visit – your home, your office, that coffee shop you hit every morning, your kid’s school, the gym – creates a digital fingerprint. Researchers have known for years that just a few data points can often identify a person. The problem was that doing this tracing, called re-identification (the process of linking anonymous data back to a specific individual), took serious manual effort. It required skilled human analysts, often spending hours or days per target. That kept the scale of the threat pretty small.
Enter agentic AI. This is AI that doesn’t just respond to prompts. An AI agent (a program that uses an advanced AI model, like OpenAI’s GPT-4, to plan and execute tasks without direct human supervision) can act like a highly efficient digital detective. It can access the open internet, browse social media, and cross-reference public records. It can connect the dots between your “anonymous” location trail and your actual identity.
Imagine a private investigator, but instead of interviewing people and staking out buildings, this investigator lives entirely online. It has access to every public record, every social media post, every news article. It can sift through mountains of information in seconds. This agent doesn’t need coffee breaks. It doesn’t get tired. It just keeps working, methodically linking pieces of information. This confused me for years: how could an AI do anything beyond typing? The trick is that these agents are programmed with a goal and given access to tools (like a web browser, a database search, or an API that connects to other services). They decide which tool to use and when, to move closer to their objective.
Here’s a simplified look at how an AI agent might work to re-identify someone:
*The AI agent analyzes mobility microdata, infers home and work addresses, searches the open web and social media, then resolves to candidate identities.*The numbers are sobering
This isn’t just theoretical. A recent feasibility study, “Agentic AI-Powered Re-Identification: An Emerging, Scalable Threat to Mobility Microdata Privacy,” by Oscar Thees, Roman Müller, and Matthias Templ, submitted to arXiv on June 26, 2026, demonstrated this capability in a real-world setting. They built an end-to-end pipeline using large language model agents (AI models like those powering ChatGPT, but designed to take actions) that could autonomously search the web, cross-reference public records and social media, and link raw location data to specific people. All without any human stepping in.
The results are stark. The researchers evaluated their system using simulated location data, focusing on high-risk scenarios. Their agentic AI successfully re-identified 18 of the 25 individuals who were, in principle, re-identifiable from the data. That’s a success rate of 72% for those cases. Looking at all 43 cases they tested, the system achieved a re-identification rate of 41.9%.
This level of success, achieved autonomously, means the cost and effort of re-identification have plummeted. What used to take skilled human analysts hours or days now takes minutes and dollars per target. Think about that for a moment. This makes large-scale privacy breaches not just possible, but cheap.
*The agentic AI successfully re-identified 72% of re-identifiable individuals and 41.9% of all cases overall, a rate far exceeding manual efforts.*What this means for your business
Most companies adopting AI are still thinking about chatbots and content generation. But the rise of agentic AI means we need to immediately pivot our thinking to its potential for autonomous action. For any business that collects, stores, or processes mobility data, this changes everything.
The implicit assumption of “de facto anonymity” is effectively dead. Regulators, particularly those enforcing strict privacy laws like GDPR, will likely view re-identification as “reasonably likely by any means” under GDPR Recital-26. This means that data you thought was anonymous might now be considered personal data, triggering a whole new set of compliance obligations.
The risk profile has shifted dramatically. A data breach involving “anonymous” location data is no longer just a data breach; it’s a potential re-identification event with significant legal and reputational consequences. CFOs and general counsels should be asking tough questions about how location data is collected, stored, and anonymized.
This isn’t about stopping AI. It’s about understanding its new capabilities and adapting our privacy practices to match. We need to move beyond simple anonymization techniques and explore more robust privacy-enhancing technologies. Think about minimizing data collection, using techniques like differential privacy, and constantly auditing your data handling practices against this evolving threat. The future of data privacy isn’t just about protecting names; it’s about protecting patterns.
Sources: arxiv.org