Why Hidden Device Detectors Fail Abuse Survivors

By Akhil Polamarasetty

The devices sold to help people find hidden cameras and trackers barely work, our research at UCL has found. Imagine you have left an abusive relationship and moved somewhere new, and yet your former partner seems to know things about you they should not. They repeat something you said in your living room or appear somewhere you told nobody you would be. You change your passwords and it carries on, until you begin to wonder whether something in your home is watching you, or whether, as you have been told so many times, you are imagining it.

This is what technology-facilitated abuse can look like from the inside: the ordinary tools of digital life turned into ways of monitoring, coercing and harassing someone. For many people it is now part of how stalking works. In England and Wales, around one in five stalking victims in the year to March 2024 were stalked by a partner or ex-partner and 42% said online methods were involved, while colleagues at UCL found cyberstalking growing faster than its offline counterpart. Most of this abuse runs through shared accounts and the phones and smart devices already in the house. Hidden devices sit at its far end: cameras built into clocks, picture frames and USB chargers, and Bluetooth trackers slipped into a toy or a car, sold through mainstream online retailers and sometimes openly marketed for catching a ‘cheating partner’.

A growing parallel market sells handheld detectors and phone apps that promise to find these devices. Coercive control can leave people living on constant alert, and gaslighting works by making them doubt their own perceptions, so being able to rule out a hidden device can bring peace of mind, and matter as much as finding and removing one. As one domestic abuse advocate told us, “We need to be ruling things out.” The question that drove our study at UCL was whether these tools can really do that.

Hard to Use and Emotionally Taxing

We gathered 19 detectors, from handheld bug detectors to phone apps that scan for trackers, wireless networks and camera lenses. We then hid nine devices, including cameras and trackers, in a mock living room setup. Thirty-four people, including police officers, network specialists, domestic abuse advocates, tech abuse researchers, and members of the public, searched it first with their eyes and then with three different tools. 

In the first round, people missed two-thirds of the devices when looking for them manually. A detector only helped them marginally, leaving them missing close to three in five. Most participants followed their instincts towards suspicious objects like clocks, toys and crowded shelves while walking straight past everyday objects like USB chargers and sockets. Instead of widening the search, the detectors mostly encouraged people to recheck the places they already suspected. Those who worked through the room methodically tended to find more, and experienced participants who had familiarity with surveillance devices, detectors or networking held a slight advantage due to prior knowledge about technical cues and likely objects.

What stayed with us most however was how the search made people feel, even though they knew nothing was at stake. One participant kept returning to an innocuous spot as the detectors beeped inconsistently, until they felt “crazy” and could no longer tell whether the fault lay with them or the device. When a tool stayed quiet, people tended to assume the area was safe and move on. Women participants more often blamed themselves for confusing results while men more often blamed the tool, a pattern our sample of 21 women and 13 men cannot confirm statistically, but one that is hard to ignore when the person holding the detector may have spent years being told she is the problem.

Actionable and Interpretable Feedback Helps

The difference between outcomes of various tools had little to do with how much they could sense. Most detectors report signal strength through beeps, lights or numbers, and leave the user to work out what has been found, where it is and whether it matters. Handheld detectors that did well in a lab setting shielded from radio signals held no measurable advantage in the room, and neither did the more expensive devices we tested.

The tools that stood out were Bluetooth tracker apps such as ‘AirGuard’. They could only find Bluetooth trackers, not devices like cameras, but they showed people a list of what Bluetooth tracking devices were nearby, and if an unknown tracker was compatible, they could make it  play a sound. Consequently, people missed only about a fifth of Bluetooth trackers, and had fewer false alarms. However, a list alone was not enough, because network scanning apps like ‘nRF-Connect’ also list every wireless device they can detect nearby, but they show each one as a string of codes and numbers rather than a name, and give no way to find where it is. People did far worse with them.

A detector, then, should be designed as an aid to decision-making. We recommend breaking a search down into five stages: detecting a signal, making sense of it, finding its source, confirming what it is, and deciding what is safe to do next. Most of the tools we tested only performed the first step. Designing for all five also means treating emotional wellbeing as a requirement, with discreet alerts and honest language that says, “no devices detected” rather than “no devices present”.

Role of Regulation and Safety-By-Design

Improved design and usability of detectors will only take us part of the way. The trackers in our study could be found largely because they are built to be found. They broadcast a signal phones can pick up and can be made to play a sound, features the IETF’s DULT working group is now turning into a standard for all trackers. Hidden cameras are built to do the opposite, so however good the detectors become, the burden to find them remains with the person being watched.

That should lead us to ask why disguised cameras are sold so freely in the first place. They are usually marketed as home security, yet much of the protection comes from people knowing a camera is there. Cameras are thought to deter crime partly by raising the perceived risk of being caught, and a camera nobody can see raises no such risk. Even the reassurance of visible cameras is less certain than we tend to assume, and a hidden one can reassure only its owner. For everyone else in the room it does the reverse, turning a home into a place where nobody can be sure who is watching.

Some people have legitimate reasons to record in secret, including domestic abuse survivors and journalists gathering evidence. Those use cases deserve to be weighed properly, but regulators should ask of these products what the law already asks of covert filming at work: is it necessary, could a less intrusive option such as a visible camera do the job, and is it proportionate to the harm done to those recorded? Until someone asks, survivors will be left searching their own homes for something designed not to be found.

Akhil Polamarasetty is a PhD candidate in the CDT in Cybersecurity and the Gender & Tech Research Lab at University College London. His research examines technology-facilitated abuse within internet-connected environments, focusing on the usability and safer design of technical surveillance countermeasures to better support survivors of interpersonal abuse.

Hidden Device Research Heads to IETF 127

Akhil Polamarasetty, author of this week’s lead story, will present the research at IETF 127 in San Francisco on Monday, November 16th , in the Human Rights Protocol Considerations (HRPC) research group session co-chaired by IX’s Mallory Knodel. HRPC researches whether standards and protocols can enable, strengthen or threaten human rights, and works to increase awareness in both the human rights community and the technical community. Akhil will discuss the implications of hidden cameras for safety and privacy, and explore how existing RFCs might relate to these issues or where additional work may be needed.

Full agenda. Stream details for Monday, 14:00 PST (22:00 UTC).

Want to appear here? Sponsor a newsletter.

This Week's Links