Theta Lake has been granted United States patent: US 12,464,032, System and Method for Visual Identification of Displayed Applications in Electronic Communications, and it addresses one of the most visually obvious yet technically overlooked risks in modern workplace collaboration:the applications that appear on screen during a screen share.
The patent intersects with and supports a broader portfolio of patents that together define our approach video compliance and oversight of digital communications: From our foundational patent covering context-based policy detection across what is spoken, shown, and shared in video communications, to patents covering participant disambiguation and AI-assisted review workflows,this patent sharpens the picture by identifying exactly which tools were visible on screen.
The Risk Hidden in Plain Sight
Screen sharing is one of the most commonly used features across collaboration platforms like Zoom, Microsoft Teams, and Webex. It is also one of the least supervised. When an employee shares their screen during a meeting, everything visible to them is potentially visible to every participant, and in recorded sessions, to anyone who reviews the recording afterward.
That might include a CRM window showing customer data. A spreadsheet with salary or financial information. An HR system, a development environment, or an email client with sensitive correspondence open in the background. These aren’t hypothetical scenarios! They happen every day, often without the person sharing their screen even realizing it.
Image attachments in emails and chat conversations are also in scope. A sensitive application doesn’t have to appear in a recorded video meeting to create risk – a screenshot pasted into a Slack message or attached to an email can expose exactly the same data, and deserves the same scrutiny.
Until now, the only way to catch these moments was for a compliance reviewer to watch recordings manually and hope they spotted the relevant frame. That approach doesn’t scale.
What the Patent Covers
Patent 12,464,032 describes an AI-based system that analyzes the visual content of collaboration sessions: screen shares, webcam feeds, whiteboards, are automatically identified when specific applications appear on screen.
The technical foundation is the concept of an app fingerprint: a signature built from the text and visual features characteristic of a given application’s interface. Textual clues can include menu item names, toolbar labels, tooltips, or URLs; visual clues might include logos, buttons, fields, or structural formatting like the grid layout of a spreadsheet. These fingerprints can be defined explicitly or learned automatically using AI. The machine learning classifiers are trained on a corpus of application screen images to identify the features that most reliably distinguish one application from another.
The categories of detectable applications are broad: Theta Lake’s AI can detect office productivity tools, CRM and HR platforms, financial applications, development and infrastructure environments, email clients, and more. These are exactly the tools most likely to display sensitive personally identifiable information (PII), confidential corporate data, or other material that organizations have a legitimate interest in monitoring.
Detection isn’t limited to analysis of video frames either. The machine learning system can draw on additional signals: If a participant mentions in chat or audio that they’re about to share a particular application, that context can inform the detection.
Consistent, Scalable Oversight
The value of automated application detection is consistency. A human reviewer watching hours of recorded meetings will inevitably miss things — fatigue, volume, and the sheer pace of modern communication make that unavoidable. An automated AI system applies the same detection logic to every frame of every session, every time.
For organizations operating under supervision and recordkeeping obligations, that consistency matters. It means that the question “was a sensitive application visible during this session?” has a reliable, auditable answer, and not one that depends on whether a reviewer happened to be watching closely at the right moment.
