Sports piracy: who’s pulling ahead in the AI arms race

Piracy of sports streaming is rampant. So much so that an ESPN reporter was accused of watching an illegal streaming site when he posted a comment relating to a recent NFL game on social media.

Meanwhile, having paid €400m a season for rights to cover France’s top football division, Ligue 1, sports streamer DAZN set a target of 1.5 million subscribers only to find that in the first week of its broadcasts around 200,000 people were illicitly streaming its coverage.

“Live is the most valuable content to try and monetise with a pirate business model” Werner Strydom, Irdeto

Cybercrime is endemic according to anti-piracy solutions vendor Synamedia which suggests the sports industry is missing out on $28bn a year as a result. Even that figure, calculated in league with Ampere Analysis, accepts that a hardcore 26% of viewers will never pay.

“Piracy is exacerbated by the fragmented content market,” says Tim Pearson, Product, Solution and Partner Marketing Leader at Nagra. “There’s anecdotal consumer feedback that says ‘I’ve paid for two or three services, I can’t pay for any more so I’ll access this one through a much cheaper service’. The problem is that many consumers don’t realise that they are buying into a criminal enterprise that is probably harvesting data from them as well.”

Werner Strydom, Head of Advanced Technology and Innovation, Irdeto, says: “There is more live streaming piracy than in the past because live is the most valuable content to try and monetise with a pirate business model.”

AI sword

AI is escalating the problem. For example, Generative AI is accelerating the process by which pirates can create teasers, clip content and publish on social media to drive audiences to their platform.

Pearson says: “Pirates run sophisticated marketing organisations and AI is making their fake content look as good as if it were created by someone like the BBC. When content looks this professional, it dupes the consumer into thinking that the site must be legitimate.”

From a forensics perspective, the use of AI by pirates is a genuine concern. High-value content like a Champions League football match is embedded with a watermark that is so subtle that pirates find it difficult to determine whether it’s present or not, let alone remove it.

“Usually, [pirates] do a lot of content manipulation in the hope that if there’s a watermark present, it damages it to such a degree that you can’t read it again,” says Strydom. “With AI they probably no longer have to guess; they will know for a fact.”

Similar risks are now occurring with upscaling. “Even if pirates have stolen a crappy SD version, they can upscale it to HD. That used to take a lot of processing and video editing skills but AI has made it a lot easier.”

That upscaling process can also scrub away the watermark.

AI shield

“Many operators are willing to accept a certain degree of piracy because they don’t want to create too much disruption” Werner Strydom, Irdeto

The shield side of the picture is that AI is also accelerating defensive capabilities. Andy Haynes, SVP of Engineering at Friend MTS, an anti-piracy provider which counts UEFA among its clients, says: “AI is a really hot topic. There’s a lot of misinformation going around about how vulnerable the industry might be to AI, but it can also be extremely useful. The real value for AI is going to be in a lot of small things that help us work more efficiently, rather than one big system that just fixes everything for you.”

He reports: “We’ve seen cases where people have got a TV on in the background and they’re effectively doing the commentary with the live broadcast. It’s not the same content as was originally broadcast but we can start to use AI to detect that and then investigate further.”

The power of AI mostly comes down to automating what used to be an extremely manual workflow. A lot of metadata gets added manually based on human judgement and with human interpretation of the results. The latest generations of AI are making it a lot easier to automate those processes.

“It’s not completely taking the human out of the loop, and I doubt that will ever be the case,” says Strydom. “But what it certainly does is make it possible for us to extend the scope of our [web] crawling to look for piracy and to process a much larger quantity of potential piracy candidates.”

Dealing with live piracy requires responding to an illegal stream within minutes of an event having started, and according to Pearson, this is a major area where AI can help in triggering and accelerating workflows automatically.

“If an algorithm detects a watermark or a fingerprint of content that’s distributed illegally, the model is also smart enough to be able to react and deal with it,” he says. “You can scan for a lot more patterns and do a lot more pattern matching with AI than you can do conventionally.”

For a popular live event, there may be tens of thousands of potential streams that need to be investigated, but only a fraction are relevant to the actual event requiring protection. “Step one is differentiating between what is in scope and what is out of scope,” says Strydom. “Using AI to make a judgement call about whether something is tennis or football or some other sport is an established capability.”

On top of that, logo recognition can be added to filter out legitimate candidates. The next step might be to issue takedown notices to the few streams that are left. There are, however, risks of scooping up legal streams in the trawl.

“A lot of operators are not entirely comfortable with takedowns being completely automated,” Strydom continues. “Instead, they may try to prioritise known pirates who are causing the most harm from a brand or revenue perspective, and not worry too much about the smaller guys. Many operators are willing to accept a certain degree of piracy because they don’t want to create too much disruption for the legitimate customer base.”

In Italy, anti-piracy platform Piracy Shield – which has been in operation since the beginning of this year and is managed by the nation’s media regulator to protect sports rights on behalf of Prime Video, DAZN, and Sky – has managed to take down legitimate providers on more than one occasion.

“Clearly that’s not a good thing,” Haynes says. “We don’t want that to happen [to us]. We have a certain amount of reticence about using AI in that decision-making process for the very reason that if it gets it wrong, the negative consequences are pretty staggering.”

Haynes says police forces are similarly hesitant about using AI in pre-emptive decision making. “They can use it for pattern analysis and to predict where things might happen, but it’s usually a very robust process when it comes to enforcement because you have to be able to stand by how you’ve reached a conclusion. It’s not good enough to just say ‘That’s what the model spat out’. You’ve got to have the actual evidential chain.”

Minority report

One of the most compelling ways AI could transform the battle against sports piracy is through advanced content recognition and detection systems. Could algorithms predict potential instances of piracy before they occur?

Haynes is sceptical. “I wouldn’t like to say ‘never’, but I don’t think you’d need to detect it before it appears so much as you just need to find it sooner. Especially for boxing matches or events that can be over very quickly, the time taken to respond is hugely important.

“You can assume people will be trying to pirate an event and you could probably use some AI behavioural analysis to judge where to look. You don’t necessarily need to recover the stolen goods, but you need to have enough evidence to suggest that someone is committing a crime. We have techniques in place in certain areas that look for indicators of piracy to support that evidential chain,” he says.

Irdeto’s Strydom says it is possible to predict piracy based on subtle signals that have been accumulated as data sets, as a result of interactions by customers with call centres and conditional access broadcast systems.

“The patterns are so subtle that a human probably can’t see anything wrong with it but if the data set were large enough and you train a model on ‘normal behaviour’ versus abnormal behaviour, it may be able to filter for possible pirates. Whether you’d go a step further and pre-empt piracy before it happens, I’m not sure. It sounds a little sci-fi, even a little scary. We’re not experimenting with anything in that area right now.”

Disrupt to desist

“It’s not necessary to always kill the stream, but to make it such a bad experience that subscribers are going to give up trying cheap versions that don’t work” Tim Pearson, Nagra

Disruption is sometimes better than cessation, especially when blocking streams becomes a game of whack-a-mole. “Shut down a pirate stream, another one will open up. Whereas if you disrupt the experience, it is less easy for the pirate to monitor,” Pearson says. “That’s where AI can really disrupt the operation. For example, you could put up an overlay on the illegal stream to tell viewers it’s not authentic but they can scan a QR code and watch the rest of the game legitimately.”

Various other counter measures can be deployed using AI that will disrupt the viewing experience. “Ultimately, it’s not necessary to always kill the stream, but to make it such a bad experience that subscribers are going to give up trying cheap versions that don’t work and will then convert back to paying subscribers.”

AI-driven dynamic pricing models

Another possible deterrent is to alter the price of legitimate streams depending on precise consumer analytics based on their preferences and local market.

“It’s definitely possible,” says Pearson. “In fact, one of our products (Insight Negotiation Agent) uses AI in a virtual call centre agent, so if a customer thinks they’re paying too much for sports they could re-negotiate a new price within the guardrail set by the operators, via the virtual customer agent.

“That technology is here. Dynamic pricing on a match day is also possible. However, you’ve got to be careful that your core paid-up subscribers don’t lose out to people who are paying a third less after converting from a pirate service. It’s a balance, and tricky to do on a match day if you’re working on a pay-per-view basis.”

After DAZN had its nose bloodied on the first weekend of its Ligue 1 coverage, it offered a temporary discount on its monthly subscription fees (from €39.99 to €19.99) in an attempt to address pricing concerns. According to a survey from French market research and polling company Odoxa, Nearly two-thirds (65%) of French football fans believe the cost of a subscription to DAZN will encourage more illegal streaming of Ligue 1.

“The price was enough to drive the illicit behaviour,” says Pearson. “Once DAZN reduced the price and deployed a load of counter-piracy measures they saw an improvement.”

No AI silver bullet

“Now that they can quantify losses due to piracy in cold financial terms, we’re seeing a change in anti-piracy attitudes” Werner Strydom, Irdeto

Experts emphasise that tackling the criminals requires a belt-and-braces and straitjacket approach.

“There’s always something new coming along that will thwart whatever model you’ve already built,” Haynes says. “Pirates try to keep in business by avoiding detection, so we find that some of the things we’ve done in the past don’t always work the second time around.”

Nagra research quotes an operator of a pirate organisation whose candid response was to call it a game of ‘who can outwit the other the quickest’. “Just because pirates are using AI, that doesn’t mean that the industry isn’t using AI,” says Pearson. “So for every benefit the pirate gets, the industry also gets another one.”

There are no signs that cybercriminals are using AI to break the cryptography associated with the digital rights management (DRM) system, yet. Irdeto’s Strydom calls it “an arms race”, but also that there does seem to be greater awareness among clients that security is no longer something that can be skimped on.

“For a long time, anti-piracy has been seen as a ‘nice to have’, not an imperative; almost as if operators were asking why they should police the net,” he says. “Now that they can quantify losses due to piracy in cold financial terms, we’re seeing a change in anti-piracy attitudes.

“For years everybody talked about multi-DRM, but that is now just a hygiene factor. Operators are learning they need a lot more than that.”