ORT is passive leakage rather than malware-driven exfiltration. During a video call, light from text displayed on a user's screen can reflect from the outer surface of their eyeglasses into the webcam. The remote participant receives that reflection as part of the ordinary video stream even though the screen itself is not being shared. The challenge is…
Inspect the ClaimsCase · DiggingBeagle record
ORT recovered reflected on-screen text from eyeglasses in video calls
ORT research automated extraction of text reflected from eyeglass surfaces captured by webcams during video conferencing, reducing the effort required for screen-peeking attacks.
Scope
Passive optical leakage from camera-visible reflections. It does not require malware on the viewed computer but depends on camera quality, geometry, text size, lighting and eyeglasses.
Assessment profile
Assessment method- Setting
- research demonstration
- Exploitation
- demonstrated
- Evidence
- Grade B
- AI role
- WITH AI
Basis and provenance
The recorded IEEE abstract supports an automated optical/OCR research technique; no named victim monitoring incident is established. Assessment is a desk review of the retained canonical Claims and cited Source metadata at their recorded cutoff, not a new external verification. Impact remains ungraded.
Assessed Sep 24, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-ORT-EVAL · CLM-ORT-MECHANISM · ORT: Unintended Text Recognition from Eyeglass Reflections in Video Conferencing Environments
At a glance
ORT shows that camera privacy extends beyond the objects intentionally framed by the user. A participant can leak screen information through reflections even when screen sharing is disabled. The practical risk is conditional rather than universal, but improving webcams and automated recognition lower the effort required to exploit favorable viewing…
Read the implicationsThe public results do not establish reliable OCR for arbitrary small text, every eyeglass type or every conferencing setup. Recognition quality depends on lighting, text size, resolution, compression, head movement, camera angle and lens properties. The case demonstrates information leakage potential, not evidence that a named victim was covertly monitored…
Limits and uncertaintyFull account
ORT is passive leakage rather than malware-driven exfiltration. During a video call, light from text displayed on a user's screen can reflect from the outer surface of their eyeglasses into the webcam. The remote participant receives that reflection as part of the ordinary video stream even though the screen itself is not being shared.
The challenge is that the reflected text is small, warped, low-contrast and mixed with glare and motion. ORT addresses that with reflection-aware preprocessing plus scene-text detection and recognition models, reducing the amount of manual inspection needed to extract useful text. The physical reason is simple: specular reflection preserves some spatial structure from the display, and improved image processing can recover information that a human observer might not notice in the raw frame.
The result does not imply universal readability of arbitrary screen content. Camera resolution, eyeglass geometry, head pose, text size, lighting, compression and lens properties all affect what survives the optical path.
Mechanism and trust boundary
- 01
Display sensitive text during a video call
The victim views on-screen content while their webcam is active; no malware or screen sharing is required.
Boundary: private display / local user
- 02
Reflect display light from eyeglasses
A portion of the screen's optical signal is redirected by the lens surface toward the camera.
Boundary: display emission / reflective surface
- 03
Capture the reflection in the video stream
The webcam records the eyeglass region and sends it through the normal conferencing path.
Boundary: local optical scene / remote video participant
- 04
Preprocess distorted reflection frames
Image processing compensates for blur, contrast loss, warping and other reflection artifacts.
Boundary: raw video / recoverable text image
- 05
Detect and recognize text
Scene-text detection and OCR convert the recovered reflection into machine-readable content.
Boundary: processed pixels / extracted information
Timeline
- Feb 24, 2026Event type unspecified
ORT record added to IEEE Xplore
IEEE Xplore added the ACSAC 2025 paper describing the automated reflected-text recognition technique.
Claims & evidence
CLM-ORT-EVALThe authors' Zoom experiments used a 1080p webcam and found that 25-30 mm text retained over 80% recognition accuracy in most tested environments. Recognition fell sharply below 20 mm; 15 mm text was almost never recognized outside the bright-room, white-screen condition. These results concern prepared reflection tests, not arbitrary screen text.supported
Basis: reported finding
- supportsORT: Unintended Text Recognition from Eyeglass Reflections in Video Conferencing Environmentsrole not specified
Authors' ACSAC 2025 presentation, slides 13-16: setup, word-level accuracy and text-size/lighting results (https://www.acsac.org/2025/files/web/acsac2025-290-kim.pdf).
CLM-ORT-MECHANISMORT uses preprocessing and scene-text detection/recognition to extract text reflected in eyeglasses from video-conferencing imagery.supported
Basis: reported finding
- supportsORT: Unintended Text Recognition from Eyeglass Reflections in Video Conferencing Environmentsrole not specified
IEEE abstract
Implications
ORT shows that camera privacy extends beyond the objects intentionally framed by the user. A participant can leak screen information through reflections even when screen sharing is disabled. The practical risk is conditional rather than universal, but improving webcams and automated recognition lower the effort required to exploit favorable viewing conditions.
Controls and mitigations
- Blur or mask eyeglass regions when sensitive screen content is present and the conferencing workflow can tolerate it.
- Disable or minimize camera exposure for tasks that display highly sensitive text when video is not required.
- Adjust display position, camera angle, lighting or eyewear to reduce useful reflections instead of assuming the remote party can see only the intended frame.
- Apply least-privilege camera policies in environments where reflected information is part of the threat model.
Unknowns and contradictions
- The public results do not establish reliable OCR for arbitrary small text, every eyeglass type or every conferencing setup.
- Recognition quality depends on lighting, text size, resolution, compression, head movement, camera angle and lens properties.
- The case demonstrates information leakage potential, not evidence that a named victim was covertly monitored in the wild.
Sources and citation
Material revision history
- Sep 19, 2026 · Published version · first publication · revision 15
- Sep 19, 2026 · Published version · material revision · revision 20Public fields changed: attack_path, implications, mitigations, narrative, timeline, unknowns
- Sep 25, 2026 · Published version · material revision · revision 36Public fields changed: assessment_profile
- Oct 8, 2026 · Published version · material revision · revision 100Public fields changed: ai_roles, claims