As reported by Music Business Worldwide, Universal Music Group’s patent-filing arm has published blueprints for an AI-powered copyright enforcement system that would not just detect music infringement across the internet but issue legal correspondence directly to the people responsible, with human lawyers brought in only as an escalation tier. The filings represent the most detailed public window yet into how the world’s largest music company intends to use AI as a weapon against AI.
The two patent applications, bearing publication numbers US 20260044581 A1 and 20260044583 A1, were published by the US Patent and Trademark Office on February 12, 2026. Both filings were submitted by Music IP Holdings, the entity formed through UMG’s partnership with IP asset management firm Liquidax Capital, with Liquidax founder Daniel Drolet listed as sole inventor. Billboard Philippines They describe a sprawling end-to-end “media rights platform” designed to sit between rightsholders, generative AI systems, and the end users who prompt those systems to create derivative works from copyrighted music.
UMG announced its strategic partnership with Liquidax in July 2025 to accelerate the development of music-related AI patents, forming Music IP Holdings to license the technology into the global marketplace. By November 2025, MIH had opened a headquarters on Nashville’s Music Row and disclosed that it held more than 60 protected innovations with numerous additional technology families and portfolios under development. Art Threat The two filings examined by MBW appear to be part of that broader portfolio and represent a significant escalation in ambition beyond defensive IP protection.
The most consequential component described in the filings is what they call an “LLM Agent Copyright Crawler,” a system designed to sample content streams from the open web, detect digital watermarks embedded in audio and images using machine learning, and cross-reference that material against IP licenses currently in force. When the crawler identifies a mismatch between what is being distributed and what has been licensed, the filings describe it as capable of connecting directly with the source, proposing licensing terms, thanking users who have “properly licensed usage,” and, critically, sending “one or more cease-and-desist letters to user or streamer.” Human legal intervention is flagged as an escalation option, not the default. That is, functionally, a blueprint for fully automated copyright enforcement at internet scale.
At the front end of the platform sits a “Copyright Licensing Chatbot” described as a plugin that integrates directly into existing large language model platforms. The filings explicitly name ChatGPT and Google’s Bard, and also reference Anthropic’s Claude, Google’s LaMDA and PaLM, Hugging Face’s BLOOM, Nvidia’s NeMo, and others. The bot is designed to interrogate potential licensees about commercial versus non-commercial use, timeframe, and geographical scope before clearing or escalating the request. Risk profiling is handled by machine learning models trained on historical licensing data and infringement patterns. Low-risk, non-commercial queries can be resolved within the conversation. High-risk requests escalate to human review.
The platform also includes an AI Modeling System that uses machine learning trained on “historical licensing decisions, legal precedents, and copyright holder behavior patterns” to predict whether a rightsholder would approve a derivative work request before it is ever created. Derivative work creation is then approved or declined based on that prediction, effectively inserting a computational proxy for the artist’s judgment into the licensing pipeline. Underpinning the whole system is a digital watermarking and fingerprinting engine designed to identify which specific AI model created a given derivative work, described in the filings as detecting “GAI-MC” fingerprints, defined as “unique byte sequences generated by specific AI model architectures.” The system is designed to calculate “how many derivatives were created, how much money is owed, how much value was used to train the LLMs in order to create the derivatives.”
Perhaps the most commercially revealing component is a dynamic pricing engine that adjusts licensing costs in real time based on demand, seasonality, and market conditions, using reinforcement learning to continuously optimize pricing based on conversion rates, customer lifetime value, and market penetration goals. The filings describe personalized subscription plans, bundle offerings, loyalty rewards, and tiered pricing for AI music licensees, a model that looks considerably more like demand-based dynamic pricing from the live ticketing industry than traditional flat-rate music licensing.
The filings are patent applications, not shipped products, and patent claims routinely describe broader territory than what gets built. Neither UMG nor Music IP Holdings has publicly disclosed commercial plans for the technology. But as a statement of strategic direction, the filings are unambiguous. In February 2026, Spotify Co-CEO Gustav Söderström said the streaming giant’s technology to let fans make AI-generated remixes and covers was “ready,” but that “the absence of a rights framework” was holding things up. Art Threat UMG appears to be building exactly that rights framework, on its own terms, at a level of automation that would fundamentally change how copyright enforcement operates in the AI era. For AI firms already fighting label lawsuits over training data, the direction of travel is clear enough.
