Industry-First Innovation
This week marks four years since the United States Patent and Trademark Office granted Theta Lake US 11,190,733 B1, a utility patent covering systems and methods for applying context-based policies to video communication content.
The patent formalized technology we’ve built to solve one of the hardest problems in modern compliance: applying business-aware, regulatory policies to video and rich media. It addresses the challenge of detecting and handling security and compliance risks in what is shared, shown, spoken, or typed across a diversity of communication platforms. Solving these problems with deep technical experience and innovative technology was the reason Theta Lake was founded.
Issued on 30 November 2021, the patent positioned us to support firms as video adoption accelerated during the Covid-19 pandemic, across platforms like Zoom, Microsoft Teams and Cisco Webex. While video humanized digital communication and strengthened engagement, it also introduced regulatory, operational, and security considerations that legacy compliance tools were not designed to manage. It opened up an avenue for data leakage and misconduct through screenshares and documents shown on camera, with misconduct appearing visually, verbally, or textually.
No compliance team could manually review the scale of video being generated. Our early insight and deep AI technology helped financial services firms to implement effective oversight of video for governance, security and regulatory compliance. That oversight remains essential today as video continues to be a core communication channel in the modern workplace.
Still the Only Video Supervision Solution of Its Kind
From the outset, Theta Lake chose not to retrofit email and audio approaches onto video. Instead, we developed new technology built specifically for multi-modal communication. Four years later, the core innovations in the patent remain unique. Theta Lake is still the only vendor in the digital communications governance and archiving (DCGA) market able to capture and supervise video content and all its components including audio, screenshares, on-screen text, visual elements, slide decks, and in-meeting chat.
Synthesising Voice, Video, and Text: How the Technology Works
Theta Lake’s platform uses machine learning algorithms for audio, image, and natural language processing to analyze what was spoken, shown, and shared during video communications to identify potential regulatory compliance, privacy, and security risks. The patent describes the multi-stage process that compares video content against pre-defined policy sets to determine a risk score, and recommend or apply actions.
1. Comprehensive Content Extraction
- Video content is ingested for analysis and processing, where relevant metadata and information are extracted. This includes
- Technical metadata, such as timestamps, file information, and participants, as well as semantic information or meaning e.g. to detect instances of collusion rather than keywords or phrases that indicate collusion.
- Audio transcription: converting spoken words to text.
- Video frame analysis: Image classification, object detection and Optical Character Recognition are used to analyse presentations and documents shared on screen and detect visual elements such as credit cards, logos, sensitive documents, inappropriate objects, or offensive material.
2. Contextual Classification and Risk Scoring
Extracted content is classified against internal and external regulatory policies to detect regulatory, privacy and security risks. For example the display of confidential data, MNPI, or sensitive information like Social Security Numbers, birthdates, or email addresses through screen shares or webcams. A risk score is then assigned to each video based on the content and classification.
Multiple elements are combined to maximize accuracy and minimize false positives and false negatives. Detection methods include:
- Lexicons
- Soundalike and lookalike matching
- Sentence embeddings,
- Convolutional neural networks,
- Language models,
- Vision-language models,
- Ensemble methods on top of image or language features or both
- Proprietary fuzzy rule based matching and
- Proprietary intelligent augmentation of soundalikes and lookalikes.
3. Automated Workflows and Storage
Based on the risk score and policy requirements, automated workflows apply appropriate handling rules to ensure video content is stored appropriately, such as:
- Compression or selective capture
- Encryption
- Defined storage location
- Appropriate retention periods and other storage parameter
4. AI-Driven User Interface
Based on classifications and risk scores, the videos and associated metadata is presented to users in a structured interface for efficient review and remediation:
- Video and related communications are preserved in native format, including reactions, GIFs, and emojis.
- Each risk detection is pinpointed on the video timeline, enabling direct navigation to relevant segments.
- Reviewers can redact or remove sensitive data to prevent unnecessary exposure.
- Smart workflows automatically route content to the right supervisory team based on factors such as geography or business unit.
Enabling Compliance Teams to Identify Risk at Scale
The volume and complexity of digital content created across collaboration tools makes manual oversight of video an impossible task. Compliance teams must oversee multiple video streams, screenshares and presentations, text appearing in slides, physical documents or information visible on camera, sensitive data displayed on desktops, as well as in-meeting chat with reactions, emojis, and GIFs.
The AI-driven detections at the core of the patent, narrow vast volumes of data down to the specific moments requiring scrutiny, enabling compliance officers to focus on relevant issues rather than watching hours of video. The AI distinguishes potential risks from non-issues by evaluating the surrounding context and intent, reducing noise and improving accuracy. For example, one of the largest networks of independent wealth management firms in the United States streamlined video compliance processes, saving 3,300 hours annually.
Regulatory obligations further reinforce the need for scalable video oversight. ESMA’s Questions and Answers On MiFID II and MiFIR investor protection clarifies that electronic communications “includes amongst others video conferencing…..”, while FINRA Advertising Rule 2210 requires video communications to be fair, balanced, and not misleading. In addition, privacy obligations and governance expectations demand the detection and remediation of sensitive data or misconduct issues.
Future Commitment to Innovation
Four years later, the foundation established by our 2021 patent continues to shape the future of communications compliance. The patent captured both a technical solution to address the most complex challenges in modern supervision, as well as the early foresight of how modern communication oversight would require multi-modal, automated and context-aware analysis.
Today, our expanding portfolio of patents reflects our continued commitment to building practical, innovative, defensible technology that helps organizations govern modern communication safely, effectively and at scale.
