The upheavals that have taken place—and in some respects are still taking place—in Olympic boxing affect not only the sport’s international governing bodies. World Boxing has replaced the IBA (formerly AIBA) as the new world governing body, thereby helping to secure the survival of boxing as an Olympic sport.
The changes may soon reach deep into the “engine room” of competition operations. On June 18, 2026, World Boxing announced a strategically focused partnership with the Chinese company Xempower as its preferred technology partner. This development deserves closer examination.
Xempower Becomes a World Boxing Partner
The partnership initially concerns so-called Event Operation & Management Systems (EOMS). These are the platforms and software required or useful for running major tournaments such as world championships. Their functions primarily include registering competitors, conducting the tournament draw, scheduling and administering bouts across multiple rings, recording judges’ scores and consolidating them into bout decisions, calculating the tournament’s progress from round to round, producing result lists, assigning officials to individual events, rings and positions, and generating tournament statistics.
Farewell to Swiss Timing’s Boxpointer?
At the international competition level, Xempower could therefore gradually replace the system developed by Swiss Timing that has been familiar for decades and is often referred to as the Boxpointer. Xempower has already demonstrated the capabilities of its EOMS at several major tournaments. These have included, for example, the U19 World Championships in the United States, the U17 European Championships in Germany, the World Boxing Cups in China and Brazil, and most recently the U19 European Championships in Serbia—all of which were already conducted using Xempower’s EOMS.
Many people may not even have noticed the change from Swiss Timing to Xempower, because the standard competition documents generated by Xempower—draws, bout schedules and results—closely resemble those previously familiar from Swiss Timing. Apparently, Swiss Timing had established standards in the EOMS field with its documentation that even a competitor could not simply ignore.
Xempower’s Ambitions Extend Beyond EOMS
That, however, will probably be little consolation to Swiss Timing. World Boxing described Xempower merely as its preferred technology partner—a formulation that does not grant the Chinese company exclusivity and would presumably allow Swiss Timing to remain in use, at least for the time being. Nevertheless, the governing body’s statement can reasonably be interpreted as indicating that it sees the future with Xempower.
The objectives involved are highly ambitious and extend far beyond the provision of an EOMS. In its statement of June 18, the world governing body explicitly said that Xempower would in future explore possibilities for AI-assisted tools for the training of officials—primarily judges, presumably—and for the development of scoring solutions.
IOC Called for Improvements in Officiating Performance
It is hardly surprising that judging in Olympic boxing is becoming a focus of attention. Judging is an extremely complex process, making it potentially prone to error and not always easy to understand or verify. As a result, it is also comparatively difficult to protect against external influence. For all these reasons, judging in Olympic boxing has long been subject to criticism.
A particularly low point at the international level was undoubtedly the 2016 Olympic Games in Rio de Janeiro. All AIBA referees and judges at the Rio Games—IBA was called AIBA at the time—were subsequently removed from officiating duties. A later investigation report commissioned from the renowned sports lawyer Richard McLaren referred to manipulation, but was unable to identify specific individuals responsible.
It is therefore unsurprising that, following Rio, the International Olympic Committee (IOC) repeatedly and emphatically called for fundamental improvements in officiating performance. In the increasingly serious dispute between the IBA and the IOC—which ultimately led to the IBA’s provisional exclusion in 2019 and permanent exclusion in 2023—this remained, alongside other important issues, a central concern until the end.
It is therefore understandable that the new world governing body World Boxing has placed officiating performance and judging at the top of its agenda. It will presumably want to demonstrate to the IOC that it is capable of safeguarding the integrity of competition. Incidents such as those that occurred in Rio in 2016 are not supposed to happen again under its authority in Olympic boxing.
Scoring Is Still Largely Done Without Data
Since the introduction of the 10-point must system, familiar from professional boxing, in 2013—appropriately, at the World Championships in Almaty, Kazakhstan—judges have generally no longer counted individual punches. Why would they? Under this scoring system, the winner of a round is always awarded 10 points, while the boxer who loses the round receives 9, 8 or 7 points, depending on the degree of their inferiority.
The factor that is supposed to determine the scoring is clearly specified in the rules: the number of quality blows landed within the rules is intended to determine who wins the bout. At its core, this principle has remained unchanged over the years and despite changes in governing bodies. Only when the bout cannot be clearly decided on the basis of the number of quality blows are technical and tactical considerations supposed to come into play.
In other words, Olympic boxing is actually intended to be judged primarily on quantitative grounds, much as athletics often measures times or distances, or as many team sports count goals. Yet this quantitative approach stands in a remarkable contrast to the fact that, following the introduction of the 10-point must system, meaningful quantitative data are generally no longer collected at all.
In practice, the 10-point must system has shifted judging from what is essentially an objectifiable quantitative domain into a more subjective qualitative one. In the end, judges more or less simply decide who they believe was the better boxer—and they generally do so without a robust empirical data set to support that judgment.
Experienced judges nevertheless tend to get their decisions right most of the time—but not always. Judges are not born with experience; they have to acquire it. And until they have accumulated sufficient experience, they will sometimes have to make decisions in a foggy grey area where the available evidence is ambiguous.
AI Could Potentially Assist with Analysis
It now appears that the use of AI could help improve the judging of Olympic boxing bouts. One conceivable approach would be to record the bout using cameras from multiple perspectives and have the footage analyzed by AI. Sensors could measure and transmit acceleration, velocity, angles and force, supplementing the information available to the system.
Other observation parameters could potentially also be considered, such as successful defensive actions, the boxers’ directions of movement and control of the center of the ring. Judges could then base their scoring on statistical analyses of landed punches, punches thrown, punch force and other variables.
Will AI Ultimately Make the Decisions in Practice?
The judges would retain the final say if AI merely provided them with analytical data. This is why such an approach is also referred to as AI-Assisted Judging. But the question is whether judges would actually make decisions that contradict the statistics displayed to them.
This would be particularly difficult if they themselves had no alternative data on which to rely, leaving them with nothing more than their subjective impression as a justification for a different decision.
Regaining the IOC’s Trust
World Boxing has inherited a difficult legacy, because its predecessor, the IBA, thoroughly damaged boxing’s reputation within the IOC. For some officials in Lausanne, boxing may have become something of a red flag in recent years. The lost—or, more accurately, destroyed—trust first has to be rebuilt and carefully maintained so that the new governing body’s provisional recognition can, as smoothly and soon as possible, become permanent recognition.
Against this background, it is more than understandable that World Boxing is now seriously addressing long-standing points of criticism. Since Rio 2016 at the latest, these have included criticism of officiating performance, particularly scoring, which—at least in Olympic boxing—determines the outcome of more than 90 percent of bouts. One thing is clear: incidents such as those that occurred at the 2016 Olympic Games in Rio must not happen again.
World Boxing appears to have correctly identified the central deficiency of scoring: the lack of an empirical data basis. Deciding medal positions without valid data is, given the capabilities available in the 21st century, fundamentally inappropriate and simply outdated. After all, medal positions in the 100-meter sprint are not decided by merely watching the finish line. Instead, the athletes’ times are measured to two decimal places—to an accuracy of one hundredth of a second.
AI in Other Sports
Boxing would not be alone in introducing AI-assisted judging—and it would not be a pioneer, either. Several other sports face challenges similar to those encountered in boxing: complex, action-dense movement sequences with highly individual patterns of execution have traditionally had to be assessed more or less through direct observation by officials.
- In artistic gymnastics, a Judging Support System (JSS) developed by Fujitsu and the FIG has been in use since 2019. In 2023, it was expanded from a sensor-based system to camera-based, AI-assisted technology. The system is intended to help prevent judging errors.
- Figure skating is also currently working on the introduction of AI, which is intended to identify and evaluate the techniques performed by skaters. This would give judges more time to assess the artistic aspects of a performance.
AI Turns Data into Interpretations
The use of AI is therefore expected to provide data. But it is worth taking a closer look at the term “data.” The word is generally associated with indisputable facts and therefore with a high degree of objectivity. In sports, data usually also refers to something that has been measured or counted—such as the time an athlete needs to complete a 100-meter sprint, or the distance achieved in a discus throw.
However, the data that an AI system might provide to judges in a boxing bout are fundamentally different. They do not originate from a standardized, neutral measurement system. The values generated by the AI are therefore not directly measured facts. Rather, they are assessments and classifications derived from raw data according to trained instructions. They are therefore better understood as interpretations of data.
At most, the video recordings supplied to the AI could be regarded as neutral data. They show—naturally from multiple perspectives—only what happened in the boxing ring. But in order to analyze that material, the AI must first have been trained using data and, above all, instructions concerning how that data should be interpreted.
The result depends heavily, among other things, on how much data the AI is trained on, which data are used, and which instructions are incorporated into the training process. Without training, an AI system would have little meaningful information to contribute to the analysis of a boxing bout.
This makes one important point clear: whoever controls the training data, training procedures and model specifications of such a system can exert substantial influence over its subsequent analyses. Such analyses should therefore not automatically be regarded as neutral, objective and indisputable data in the same way as, for example, a measured distance.
Whoever Trains the AI Can Exert Influence
The introduction of AI into the judging of elite sporting competitions raises a wide range of fundamental questions. The considerable influence that can potentially be exerted during the AI training process and during its subsequent operation creates, in principle, the possibility of deliberately or unintentionally steering such systems in particular directions.
What might such potential influence look like in practical terms in boxing? The following examples are not statements about Xempower’s actual systems. They are merely intended to illustrate the general governance risks that can arise when AI-assisted judging systems are used.
For example, one could imagine an AI system being trained—or being instructed during live operation—to assign greater value to attacks executed near the red corner, or alternatively the blue or white corner, than to attacks elsewhere in the ring.
“Assigning greater value” could mean, for example, that punches whose status as landed blows is not clearly established are classified by the AI as landed with an assumed 1.1‑times-higher probability when they occur in that area. This would apply equally to both boxers. However, anyone aware of the instruction could exploit the advantage by deliberately initiating attacks in areas where the AI assigns them a higher value.
Similarly, it is conceivable that targeted AI training—or instructions issued during live operation, provided that such access to the live system were technically possible and not adequately secured—could favor particular punches or combinations in the analysis.
Such factors could also be combined. They could be changed from year to year or even during an ongoing competition, provided someone had access to the system. What applied in the current year, or perhaps only to a single bout, could later be replaced by a different manipulation. Patterns would therefore be difficult to detect.
These are, of course, hypothetical scenarios involving forms of influence that appear technically possible in principle. And naturally, the same issue applies to every sport in which AI is given a role in judging. The examples above do not indicate that Xempower—or anyone else—intends to develop such a capability, already possesses one, or has ever used one. However, anyone considering AI as an aid to competition decisions should take these questions into account.
Even Subtle Manipulation Can Have an Effect
If such manipulation were carried out subtly, it would probably be barely perceptible—or perhaps completely imperceptible—to people observing a boxing bout amid the inherent noise and complexity of the action. Most likely, even experts would have difficulty recognizing it.
This would be particularly true if people assumed that information supplied by an AI system must always represent objective values simply because the system presents them electronically.
But could such subtle manipulation really determine the outcome of competitions? This needs to be considered carefully.
A substantial disparity in performance could probably only be neutralized through massive manipulation, which would at the same time be conspicuous. But conspicuous manipulation would also mean that the manipulation had failed.
Invisible Within Statistical Variation—Yet Effective Over Time
An example illustrates the principle. If a loaded die produces a “six” 50 times in 100 rolls, everyone involved will become suspicious. Statistically, a fair die would be expected to produce a six approximately 16 or 17 times in 100 rolls.
If, however, a six appears 20 or 25 times in 100 rolls, most people would still regard this as an unremarkable statistical fluctuation—a lucky streak. And indeed, even a fair die can occasionally produce 20 or 25 sixes in 100 rolls. The loaded die simply produces that outcome with a significantly higher probability.
This becomes detectable only when the sample size is sufficiently large. The more subtle the manipulation, therefore, the more likely it is to remain undetected—and remaining undetected would be the primary objective of anyone attempting to cheat at a game of dice.
The manipulation would therefore have to be calibrated so that it remained statistically inconspicuous when only a limited number of events could be observed. Such a loaded die would certainly lose several games in succession from time to time. But over a longer period, the subtly loaded die would nevertheless prevail.
This illustrates how subtle interventions can have an effect precisely because they are difficult to detect.
Applied to boxing, this would mean that weak boxers could hardly reach the top of a tournament’s medal table through restrained—and therefore inconspicuous—preferential treatment. Over the course of a tournament, they would encounter opponents whose superiority was so pronounced that even subtle, and therefore still inconspicuous, advantages would not be sufficient to help them win those bouts.
The situation could be different for strong boxing nations. If strong boxers with a hypothetical secret advantage behind them won bouts against clearly inferior opponents, the outcome would still appear consistent with their observable superiority.
As they progressed through the tournament and increasingly encountered opponents of comparable ability, such a hypothetical secret advantage might provide them with the additional two or three apparent scoring events in the AI-generated analysis that could ultimately determine a bout.
After all, the judges would have no other basis for their decision than their own vague observations and the equally vague subjective impression derived from them, apart from whatever information the AI displayed to them.
Control in the Engine Room
Does this mean that introducing AI into boxing judging ultimately achieves nothing because one potential source of harm is merely being replaced by another? Where judges could once manipulate competition decisions, could AI do so in the future?
Yes, an AI-assisted scoring system could in principle be manipulated if its training data, model parameters or operational instructions were deliberately altered. But the issue should not be reduced to that.
It is certainly the right approach to finally place scoring on an empirical foundation. AI should, in principle, be capable of analyzing processes as complex and difficult to standardize as a boxing bout.
The important caveat is that new tools and methods should not be assumed to be neutral merely because they appear technologically modern. That would perhaps be somewhat naïve.
In this context, it is worth remembering that international sporting success has been part of foreign policy in many countries around the world, and remains so. Medals bring prestige. This applies particularly to Olympic sports, which attract the attention of the global public every four years.
Whether AI could fundamentally become a potential gateway for manipulation in sporting competition decisions is likely to depend to a considerable extent on two questions:
- Who regulates and supervises the AI’s training? During training, the AI learns the sport. Sporting and technical expertise is obviously central at this stage. But even here, it may be possible to establish the foundations for potential manipulation.
- Who regulates and supervises the AI’s deployment? A trained AI system is not necessarily immutable. Depending on its technical architecture, its behavior can also be influenced through model versions, configurations or operational instructions.
Control in the engine room will therefore be crucial.
Sport would be well advised to regulate and supervise both the training of AI systems and their subsequent deployment in competition decisions in precise detail. This is particularly important when AI is used in the area where its strengths are most apparent: assessing complex situations.
Because the complexity of the situations being assessed can simultaneously make certain forms of intervention difficult to detect.
Transparency note: This text was, somewhat ironically, translated from German into English with the assistance of AI and subsequently proofread.
