Intelligence Piracy: The Match That Was Never Played

In the earlier articles in this series, we covered how, for decades, the fight against piracy has turned on one simple question: was the content copied?

It is the foundation on which much of copyright enforcement has been built. A film is copied. A television programme is rebroadcast. A football match is streamed without permission. In each case the infringement is visible, because the work itself has been reproduced.

The scale of that problem is enormous and still growing: during the 2024/25 season alone the English Premier League says it detected more than 645,000 infringing live streams and almost 900,000 unauthorised recorded clips of its matches ¹. Enforcement has kept pace, with coordinated raids such as the 2024 operation that dismantled an illegal service estimated to supply 22 million households across the UK and Europe ².

But artificial intelligence is quietly changing the question.

A football match is more than video

To a viewer, a match is entertainment. To an AI, it is data.

Every broadcast contains thousands of decisions and relationships that extend far beyond the ninety minutes on screen: player positioning, passing networks, defensive shape, pressing triggers, set-piece routines, coaching adjustments, referee behaviour, weather conditions, crowd reactions, and commentary explaining the tactical logic as it unfolds.

This is not an abstraction. Modern football is already measured at extraordinary resolution. A single match generates 3,000 to 4,000 recorded on-ball events, every pass, shot, tackle and interception, while optical tracking systems capture the position of every player and the ball several times per second, producing a continuous map of movement that never appears in the event log ³. Specialist providers such as Opta, StatsBomb, Second Spectrum and TRACAB have built entire businesses on capturing and selling this information ³.

Each inpidual match is therefore a rich source of structured intelligence. Across decades of broadcasts, those observations have become one of the largest repositories of "sporting intelligence" ever created.

A human analyst might watch hundreds of matches across a career. An AI could absorb millions.

Sporting intelligence was always a valuable asset

Sports rights have traditionally been valued according to audience size. Broadcasters pay for exclusivity because millions of people want to watch the event live, and the figures are staggering Sky and TNT Sports paid a record £6.7 billion to share domestic Premier League rights for the four seasons from 2025/26 ⁶⁷.

But viewing figures are only part of the value. The broadcast also captures the accumulated expertise of players, coaches, analysts, officials, producers and commentators: every tactical innovation, every successful formation, every coaching adjustment, and every analysis explaining why one decision worked while another failed.

A new kind of extraction

This is where the concept of intelligence piracy begins to perge from traditional copyright infringement.

Traditional piracy distributes creative works without permission. Intelligence piracy extracts the knowledge from those works and uses it to build something new. The original content becomes the training material rather than the final product. The broadcast is no longer the destination. It is the raw material.

The early legal skirmishes are already revealing how hard this is to police. In The New York Times v. OpenAI, the AI developer’s central defence is that training a model on protected text is a “transformative” and “non-expressive” analytical use — that the model learns from the material rather than reproducing it ⁸. And in the first UK judgment to grapple with generative AI training, Getty Images v Stability AI (2025), the High Court found that the model’s trained parameters do not contain or store copies of the images used to build it, and so were not “infringing copies” at all ⁹. For anyone trying to protect the intelligence inside a broadcast, that finding is a warning: the law’s traditional test — has a copy been made? — may simply return the answer no.

Learning without copying

The commercial incentive here is straightforward. A broadcaster invests billions to acquire rights, produce live coverage, and capture every tactical nuance of the game. An unlicensed AI developer could simply treat those broadcasts as free training material, building products that compete in adjacent markets without ever paying for the underlying content.

Consider an AI coaching platform trained on every professional football match played over the last twenty years. Instead of watching thousands of hours of footage, a coach could ask,

>“How would Manchester City have approached a low defensive block in 2019?”

>“Design a training session based on Jürgen Klopp’s pressing philosophy.”

>“Predict how a newly promoted side is likely to defend against Arsenal.”

That platform could be sold to amateur clubs, professional teams, academies, universities and national associations, with the AI company profiting from tactical intelligence extracted from copyrighted broadcasts without ever licensing the underlying footage.

The commercial opportunities extend well beyond coaching. In sports betting, an AI trained on decades of matches could identify patterns that humans struggle to detect, from referee tendencies and substitution timing to the impact of injuries, tactical adjustments, weather conditions and momentum swings.

Those insights could power betting syndicates, prediction services, premium subscriber platforms or fully automated wagering algorithms. In each case, the broadcast is no longer valuable because people watch it—it becomes valuable because it trains an AI capable of generating commercially valuable predictions and recommendations.

The investigation challenge

This creates a problem investigators have never really faced before.

If a pirate IPTV service is distributing live Premier League matches, investigators know what to look for. Capture the stream. Identify the server. Trace the payment infrastructure. Preserve the evidence.

But how do you investigate an AI system that has learned from thousands of broadcasts without reproducing any of them? How do you prove where that intelligence originated?

Future investigations may look very different.

Rather than tracing illegal streams, investigators may need to examine training datasets. Rather than identifying copied footage, they may analyse statistical similarities between model outputs and proprietary content. Rather than seizing servers carrying pirated broadcasts, they may seek evidence of how copyrighted material was ingested, processed and retained during model training.

The evidence itself will evolve — and, as Getty v Stability AI showed when the claimant struggled to establish where training had even taken place, so will the questions of jurisdiction and proof that surround it ⁹.

Preparing for the next decade

The next generation of piracy may not involve illegal streaming platforms at all. Instead, it may involve AI systems capable of reproducing decades of sporting knowledge without ever replaying a single match.

That presents new commercial, legal and investigative questions. How do rights holders protect intelligence rather than footage? How do they detect extraction rather than distribution? How do they license datasets rather than broadcasts? And, most importantly, how do they prove that the intelligence inside a model came from content they own?

These questions are no longer theoretical. As AI becomes ever more capable of learning from enormous collections of copyrighted material, the value of sport will extend far beyond the live broadcast.

Footnotes

1. Premier League anti-piracy figures for Season 2024/25, as set out in the Premier League’s submission to the Office of the United States Trade Representative (2025 Special 301 review). https://downloads.regulations.gov/USTR-2025-0018-0056/attachment_1.pdf

2. “Illegal £2.5 billion IPTV operation that streamed Premier League to 22 million households busted in major crackdown,” LADbible, 29 November 2024. https://www.ladbible.com/news/crime/iptv-watch-premier-league-uk-010756-20241129

3. “The Three Types of Data in Football: Event, Tracking, and Physical,” The Football Analyst. https://the-footballanalyst.com/the-three-types-of-data-in-football-event-tracking-and-physical/

4. “What Is Not Protected by Copyright Law?” Copyright Alliance (idea–expression dichotomy; TRIPS Agreement, Article 9(2)). https://copyrightalliance.org/faqs/whats-not-protected-by-copyright-law/

5. Football Dataco Ltd and Others v Yahoo! UK Ltd and Others, Court of Justice of the European Union, Case C-604/10, judgment of 1 March 2012 (EUR-Lex). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A62010CJ0604

6. “Premier League broadcast deals for 2025-2028,” Premier League. https://www.premierleague.com/en/news/3703577

7. “Premier League TV rights sold for record UK£6.7bn as Sky, TNT and Amazon secure packages,” SportsPro. https://www.sportspro.com/news/premier-league-tv-rights-deal-sky-tnt-amazon-2025-2029/

8. “Judge allows ‘New York Times’ copyright case against OpenAI to go forward,” NPR, 26 March 2025. https://www.npr.org/2025/03/26/nx-s1-5288157/new-york-times-openai-copyright-case-goes-forward

9. Getty Images (US) Inc & Others v Stability AI Ltd [2025] EWHC 2863 (Ch), High Court of England and Wales, judgment of 4 November 2025. https://www.judiciary.uk/wp-content/uploads/2025/11/Getty-Images-v-Stability-AI.pdf

10. UK Government, Report on Copyright and Artificial Intelligence (following the December 2024–February 2025 consultation), Department for Science, Innovation and Technology / Intellectual Property Office. https://www.gov.uk/government/consultations/copyright-and-artificial-intelligence