Devilray Exposes Blind Spots in FBS Detection
DAEJEON, South Korea — Governments use them to track phones. Criminals drive them through city streets and send texts that look like they came from your bank. For 20 years, researchers have been building tools to catch these fake cell towers. They have tested those tools mostly against fake towers they built themselves.
That loop has left gaps. Researchers at KAIST built a platform called Devilray that reproduces 2,592 distinct fake-base-station behaviors. They ran it against seven existing detection systems, and every one of them missed at least 192. The finding points to a larger problem with how the field tests itself: detectors have been evaluated against the wrong adversary.
An echo chamber
Fake base stations go by several names: IMSI catchers, cell-site simulators, StingRays. They pose as legitimate cell towers and get nearby phones to connect. Once a phone is connected, the fake tower can make it give up the identifier that ties it to its owner or reveal where it is. It can also cut the phone off, push it down to an older and weaker network, or send it messages that never passed through a carrier. Police and intelligence agencies have used these devices for surveillance for years. Criminals now use them for illegal surveillance and for SMS-blaster scams, in which a device carried in a car sends phishing texts to every phone nearby.
The commercial machines are expensive, and their sale and operation are tightly restricted. Universities generally cannot buy one. Carriers and investigative agencies are the ones responsible for finding fake towers in the field, but even they cannot get every model, let alone every new variant.
So researchers do the reasonable thing. They build a fake base station, study the traces it leaves, design a detector that recognizes those traces and report that it works. The next paper improves on the last one, and detection rates climb.
But the attacker in those experiments is one the defender already knows. The attack used to design a detector and the attack used to test it rest on the same assumptions. A detector that scores well may simply be recognizing its own reflection. What the field has lacked, the researchers argue, is adversarial thinking. The question should not be whether a detector catches a known attack. It should be what an attacker would do instead.
"Detection has been built around the attacks we already know," said Yongdae Kim, the KAIST professor of electrical engineering who led the work. "Real attackers don't stay there. Until now, the playing field has tilted toward them."
A standard full of open doors
Attackers have that much room because of how LTE works. When an LTE phone first connects, it does not check whether the base station is genuine. It acts on some messages before any authentication takes place, and it can be pushed down to older generations of network. These are not implementation bugs that a patch can fix. They are written into the standard, so an attacker can choose from many messages and stay fully standard-compliant.
Take the phone's permanent identifier, the main prize for a surveillance device. A fake tower can simply ask for it. That request is the move that Rayhunter, the Electronic Frontier Foundation's open-source detector, watches for. The tower can also send a rejection message that makes the phone reconnect and hand over the identifier without being asked. The standard allows both. A detector watching one door does not see the other.
"LTE is fundamentally broken," Professor Kim said. "It is full of standard-compliant messages that can be used for bad purposes."
Nobody is going to fix that. LTE's security design was settled long ago, and the standards community has moved on to 5G. Yet LTE remains the backbone of mobile service in most of the world. Many 5G networks still lean on it, and 5G phones drop back to LTE whenever 5G is unavailable, or whenever an attacker forces them to. These weaknesses will stay in the field as long as LTE does. As a result, defending against fake base stations looks less like fixing a software bug and more like an ongoing contest: a new attack appears, a detector learns to catch it, the attacker changes. That contest only works if defenders can see the moves attackers are able to make.
Thinking like the attacker
Devilray reverses the usual order. It maps the attacker first and looks at the detectors only afterward.
The team began with a commercial fake base station, a unit sold to law enforcement and intelligence agencies. It is the first such device to be analyzed in an academic study. The device could do several things that academic prototypes rarely model:
• copy the identity of a legitimate tower,
• hop between frequencies,
• run two networks at once,
• dial its power down to cover only a few meters.
The researchers then looked past that single device. They went through prior research and the 3GPP standards that govern cellular networks for other routes to the same ends. They broke a fake-base-station operation into four stages: surveying the local network, launching a fake cell, luring phones to connect, and carrying out the attack. Across those stages they identified seven dimensions an attacker can vary. Twenty-one dependency rules remove combinations that could not occur together in a real exchange. What remained were 2,592 configurations.
Every configuration was then run against real smartphones to confirm that the full procedure completed. That check matters in both directions. An attack space derived only from the specifications could include thousands of combinations no phone would actually accept. A baseline copied from a single commercial device would inherit that device's limits. Devilray is anchored in the real device and extends outward into behaviors the standard permits.
The researchers stress that 2,592 is not a count of attacks seen in the wild, and not a list of 2,592 ways to beat every detector. It is something the field has not had before: a realistic adversary, grounded in a real device and the standard itself, against which detectors can finally be tested.
Seven detectors, seven sets of blind spots
The seven detectors the team tested are publicly available and work on different principles: signaling messages, base station information, radio signals, statistical patterns and machine learning. Every one had blind spots, and most traced back to an assumption built into its design.
Some detectors key on one telltale message, so they missed attacks that reached the same goal by another route. Others flag the unusually strong signal of a cheap fake tower; a fake tower that turns its power down, as the commercial unit can, stays under their thresholds. One statistical detector watches how often phones in an area are asked to reveal their identifiers. A fake tower that collects identifiers patiently, from a fraction of the phones nearby instead of all of them, produced a pattern that looked like an ordinary bus terminal.
The harder question
The deepest blind spot is the one where the attack starts to look like a normal network. The researchers tested this directly. They retrained a machine-learning detector on the Devilray attacks it had missed. On the radio link layer (RRC) model, its detection rate rose from 51 percent to 96 percent, while its false-alarm rate on ordinary carrier traffic nearly quadrupled, from 1.4 percent to 5.3 percent. On the core network layer (NAS) model, the detection rate rose from 27 percent to 63 percent, and its false-alarm rate increased from 4.7 percent to 8.8 percent. When trained on the hardest case, that false-alarm rate reached 15 percent, particularly for the NAS layer model.
A detector that catches everything but constantly flags ordinary networks is of little use, and neither is a quiet one that misses carefully chosen attacks. The problem is not simply to maximize detection. It is to find the boundary between malicious and legitimate behavior when the standard lets an attacker move right up against it.
"The real question is whether we can reduce misses without raising false alarms," Professor Kim said.
The researchers argue that future detectors should report both numbers together. Claiming better coverage without testing against ordinary traffic gives an incomplete picture of whether a detector can actually be deployed.
An attacker for defenders
A platform that can imitate thousands of attack behaviors could also be misused, and the researchers acknowledge the risk. They share Devilray only with vetted researchers and organizations, under controlled access. What defenders have lacked is the same range of behavior to test against. With Devilray, carriers, investigators and detector developers can do that without having to buy every device on the market.
The name is a jab at the StingRay, the commercial device whose name became shorthand for the whole category. A devil ray is a much bigger relative.
The work is already reaching defenders. The team gave EFF the attack cases Rayhunter misses, and the foundation is now working on ways to detect them. The GSMA, the global association of mobile operators, invited the researchers to present Devilray to its Fraud and Security Group. There, standards and industry representatives discussed where fake-base-station detection should go next.
The study was led by KAIST in collaboration with South Korea's National Security Research Institute, Kyung Hee University, the University of Florida and Cape, a U.S. mobile network company. It will be presented in December at ACSAC 2026 in Los Angeles.
Professor Kim's group is now building a detection system of its own. No single vantage point may see enough, so it combines what phones observe, what carriers can see inside their networks and what independent radio sniffers capture. The aim is to detect fake base stations and also to locate them.
"Devilray is not meant to defeat existing detectors," Professor Kim said. "It gives defenders the same choices attackers already have, so that attack and defense can finally advance together."
Paper: "Devilray: A Reconfigurable Adversarial Baseline for Revealing Blind Spots in Fake Base Station Detection," IEEE ACSAC 2026, Los Angeles, Dec. 7–11, 2026. arXiv: https://arxiv.org/abs/2605.19232 · Project page and demo videos: https://sites.google.com/view/devilray
Access: Devilray is available to vetted researchers and organizations under controlled access. Requests: https://sites.google.com/view/devilray
Authors: Led by KAIST — Taekkyung Oh and Duckwoo Kim (co-first authors), Beomseok Oh and Yongdae Kim (KAIST); Hansung Bae and Byeongdo Hong (National Security Research Institute); CheolJun Park (Kyung Hee University); Tyler Tucker, Nathaniel Bennett and Patrick Traynor (University of Florida); Sangwook Bae (Cape).
Funding: Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government(MSIT) (No. RS-2024-00441762, Global Advanced Cybersecurity Human Resources Development).
Contact: Professor Yongdae Kim, School of Electrical Engineering, KAIST · yongdaek@kaist.ac.kr
KAIST Develops Smartphone-Based Technology to Detect Hidden Cameras
A smartphone can now be transformed into a “hidden-camera detector.” KAIST researchers have developed an AI technology that can detect hidden cameras using only a smartphone and a low-cost LED device. This new security technology enables users to protect their privacy more easily and is expected to help prevent illegal filming in everyday spaces such as hotels and short-term rentals.
KAIST (President Choongsik Bae) announced on August 30 that a research team led by Professor Jun Han of the School of Computing, in collaboration with the National University of Singapore and Singapore Management University, has developed “SweepLED,” a technology that detects hidden cameras by attaching an LED case to a smartphone.
As hidden cameras are increasingly being installed in everyday spaces such as hotels, short-term rentals, and restrooms, the need is growing for detection technology that everyday users can easily use. However, existing portable detectors require users to visually identify bright reflective spots, which can lead to false positives by mistaking reflections from metal, glass, or glossy plastic surfaces for camera lenses.
SweepLED works by keeping the smartphone camera fixed while changing only the direction of the LED illumination, then analyzing the patterns of reflected light that appear on object surfaces. Reflections from ordinary glossy objects tend to move or disappear depending on the direction of the light. In contrast, camera lenses show distinctive deformation patterns in their reflections due to their internal lens, aperture, and sensor structures.
The research team uses deep learning-based analysis to distinguish these differences in temporal reflection patterns. While conventional detection methods rely on the user’s eyes to simply look for “bright spots,” SweepLED is different in that it analyzes both the movement and shape changes of reflections across multiple lighting angles.
This enables more reliable detection of hidden camera lenses inside various everyday objects commonly found in lodging spaces, such as chargers, clocks, remote controls, and everyday objects.
The research team evaluated SweepLED on 30 objects that may be found in real-world environments and found that it achieved approximately 94% detection accuracy. It also took less than five seconds to inspect a single object.
In addition, the core components of the LED case attached to the smartphone cost less than USD 7, or about KRW 10,000, demonstrating the potential for this technology to be developed into an affordable detection tool that general users can easily access.
Professor Jun Han said, “Hidden cameras pose a serious threat to personal safety and privacy in everyday spaces,” adding, “This research is meaningful in that it combines low-cost smartphone-based hardware with AI analysis to present the possibility of a practical detection technology that even non-experts can use.”
This paper, with KAIST doctoral student Jonghyuk Yun as first author, was presented on June 20 at ACM MobiSys 2026, one of the leading international conferences in the field of mobile computing.
Paper title: Hide-and-Sweep: Detecting Concealed Cameras via LED Illumination Sweeps
https://doi.org/10.1145/3812835.3814866
Author information: Jonghyuk Yun (first author), Jaeyoung Moon, Yunseo Park, Sean Rui Xiang Tan, Byunghyun Kim, Rajesh Krishna Balan, and Professor Jun Han (corresponding author)
This research was supported by the STEAM Global Convergence Research Support Program and the Mid-Career Researcher Program of the Ministry of Science and ICT and the National Research Foundation of Korea.
KAIST Develops AI to Detect ‘Foreign-Linked Opinion Manipulation’ in 110 Million News Comments
During election seasons or major national issues, online news comment sections often become heated spaces of conflict across gender, generation, and political lines. For years, there have been persistent concerns that behind some of these conflicts may lie “foreign winds,” or interventions by foreign actors seeking to manipulate public opinion and deepen social divisions. A KAIST research team has now developed a technology that uses big data on two decades worth of news comments and artificial intelligence (AI) to precisely detect traces of such hidden influence operations.
KAIST (President Choongsik Bae) announced on the 12th of August that a joint research team led by Professor Wonjae Lee of the Graduate School of Culture Technology, Professor Meeyoung Cha of the School of Computing (Scientific director at Max Planck Institute for Security and Privacy) and Professor Alice Oh of the School of Computing, in collaboration with Professor Thorsten Holz of the Max Planck Institute, has developed an explainable AI technology that automatically detects patterns suspected of foreign-linked influence operations in online news comments and provides specific evidence for its judgments.
The organized and repeated posting of comments or content by certain actors to shape public opinion in a desired direction is known as an “online influence operation.” Existing AI-based detection technologies have had a key limitation: even when they classify certain accounts as belonging to a “blacklist,” they often fail to provide clear evidence explaining why those accounts should be considered influence-operation accounts.
To overcome this limitation, the research team used 70 foreign-linked accounts previously identified by the Institute for National Security Strategy as starting points. They then tracked groups of accounts connected to them or repeatedly commenting on the same news articles, ultimately collecting and analyzing a large-scale dataset of 110 million comments posted on Naver News over a 20-year period from 2006 to 2025.
In particular, the AI developed by the research team examines accounts through a careful three-step process. First, it checks whether there are clues suggesting that the author may be linked to a foreign source. Second, it examines whether the comment contains emotionally polarizing expressions, such as moral condemnation or blind praise. Third, it identifies which country or target the emotional framing is directed toward.
The model does not stop at simply labeling an account as suspicious. It also highlights the specific phrases in the comments that served as the basis for its judgment. The team further combined this with multidimensional behavioral-pattern analysis, including account activity frequency, account lifespan, and activity links with other suspected accounts. As a result, among approximately 4 million Naver News users, the model ultimately identified 23,998 accounts exhibiting patterns consistent with suspected public-opinion manipulation.
The analysis also revealed the more subtle strategy of these suspected accounts. Their main target was not the victory of a particular political camp, but rather the maximization of division and confrontation within Korean society.
Among the top 10 targets that drew the highest public engagement, measured through likes and other reactions, seven were prominent domestic political figures. Notably, the attacks were not concentrated on a single party or ideology. Former and current presidents, presidential candidates, and political parties from both progressive and conservative camps were targeted across the spectrum. According to the research team, this suggests a sophisticated strategy aimed not so much at supporting a particular group, but at inflaming domestic political conflict and increasing social distrust and polarization.
This study is significant because it provides data-based evidence for influence-operation activity that had previously been discussed largely in terms of suspicion, while also offering a potential defense mechanism for protecting healthy online public discourse. In the future, portal platforms and related organizations could use this technology during elections or national crises to monitor the influx of suspicious accounts in real time and prioritize the review of coordinated attacks against domestic political figures. However, the research team emphasized that the AI should not be used to block accounts indiscriminately, but rather as an explainable content-moderation tool that supports the judgment of expert reviewers.
Professor Wonjae Lee said, “By analyzing 20 years of data, we found that suspected accounts tended to use messages criticizing Korea and domestic political figures rather than directly praising foreign countries, and that these messages gained higher visibility,” adding, “This research can provide empirical criteria for when and which messages platforms and monitoring organizations should prioritize for review, especially during socially sensitive periods such as elections.”
Professor Alice Oh said, “This is a meaningful achievement in which AI precisely identified not only the surface meaning of words in massive comment datasets, but also subtle emotional patterns and organized behavioral signals intended to provoke conflict,” adding, “It can become a powerful defense system against online influence operations, which are becoming increasingly sophisticated.”
Professor Meeyoung Cha said, “This study goes beyond simple blacklist-account analysis and represents the outcome of actionable data science that addresses real-world problems and drives practical change,” adding, “In an online environment where social conflict is deepening, we hope this technology will serve as a practical tool for protecting the transparency and trustworthiness of the digital public sphere.”
This research was led by KAIST Ph.D. candidate Jaehong Kim and master’s student Hyeonseung Kim as co-first authors. The paper is scheduled to be presented at the USENIX Security Symposium 2026, one of the most prestigious conferences in the field of computer security.
Paper title: Cross-National Information Attacks: A Two-Decade Analysis of Troll Behavior in Korea,
DOI: 10.48550/arXiv.2606.22785
This research was supported by the Hyundai Motor Chung Mong-Koo Foundation, the Institute of Information & Communications Technology Planning & Evaluation, and the National Research Foundation of Korea, funded by the Ministry of Science and ICT.