Sport is buying artificial intelligence faster than it is governing it. That gap is usually described as an ethical problem. It is also a performance problem, and the performance cost arrives first. Coaches do not act on advice they cannot interrogate. Athletes do not trust systems that will not explain themselves. A club that cannot say who approved a decision cannot learn from it. Ungoverned AI in sport is not mainly dangerous because it is unsafe. It is wasteful because it goes unused.
Key findings at a glance
- Individual injury prediction does not yet work well enough to govern a career. Hamstring injury models tested across seasons returned a median area under the curve of 0.52, where 0.5 is a coin toss. The acute to chronic workload ratio adds nothing to an intercept-only model, at 0.574 against 0.5, and that is within its own training sample.
- A counted fact beats the models. Track and field athletes who completed more than 80 per cent of planned training weeks were seven times more likely to hit a performance goal, at an area under the curve of 0.72. Any programme can audit that threshold, and it outperforms the black boxes on their own published numbers.
- The law already gives athletes more than most clubs deliver. Access, portability, rectification, erasure, human intervention where automated decisions apply, and a prior impact assessment are existing rights. What is missing is an explanation on request and a record that travels with the athlete. Neither requires new law.
- A human signature is not oversight. Automation bias is well documented in clinical decision support, and its mediators are workload, task complexity and time pressure. That describes a training ground exactly. The European legislature reached the same conclusion, obliging deployers to guard against over-reliance by name.
- Sport is barely in scope, which is the problem. The word athlete appears nowhere in the European Union’s Artificial Intelligence Act, and sport appears only in its recitals. No governing body had published a binding instrument on performance AI. Sport will be governed by whatever it writes for itself.
- There is a hard line at health. A dashboard reporting what an athlete did sits outside the medical device regime. A tool applying automated reasoning to output an individual’s future injury risk has described itself into the regulatory definition, which names prediction, prognosis and injury expressly.
Why governance is a performance capability rather than a compliance chore
The argument in this paper runs against the usual order of things. Governance is normally introduced after a problem, as a brake. In performance sport it works better as an accelerator, because the constraint on using AI well is not model quality. It is trust, and trust is manufactured by explanation and accountability rather than by accuracy alone.
Consider what a coach does with a number they cannot question. The experienced ones ignore it, which wastes the investment. The inexperienced ones follow it uncritically, which is worse. Both outcomes trace back to the same missing thing, which is the ability to ask where a recommendation came from and what would change it.
Classify decisions, not technologies
Most sports AI policy starts by classifying software, and gets stuck immediately, because the same tool can be trivial in one hand and consequential in another. The paper proposes classifying the decision instead, along two axes that anybody can apply without a technical background: how serious the consequence is, and how reversible it is.
A session plan suggestion is low consequence and fully reversible. A deselection is high consequence and, in practice, irreversible for that athlete in that cycle. Anything touching health sits in its own category with its own rules. Once decisions are classified, the controls follow, and the argument about which vendor to buy becomes a much shorter conversation.
The five controls
The paper sets out five controls that a federation, club or institute can adopt without creating a new department.
- Data rights and athlete agency. The athlete’s record travels with them, and they can see it, correct it and take it.
- Evidence and validation. Every number a practitioner sees carries a stated evidence grade, so that a well-supported figure and a plausible guess are not displayed identically.
- Explainability a coach can use. Provenance, uncertainty, assumptions and real alternatives, rather than a confidence percentage with nothing behind it.
- Bias, inclusion and uneven evidence. The evidence base is thinner for women, for para athletes, and for most events outside the sprints and the marathon, and a system that hides that is misleading its users.
- Human accountability. Name who recommends, who approves, who acts, who reviews and who answers, and never let those collapse into one person.
The habit that carries all five is writing decisions down, with their reasons, on the day they are made. A decision record costs a minute at the time and cannot be reconstructed afterwards at any price.
What follows for a board
None of this is expensive. The register fits on one page. The three meetings it needs already exist under other names. The paper sets out 12 procurement questions to ask before signing anything, and a 100-day sequence for adoption.
What the discipline buys is the ability to learn from a season rather than merely remember it, and the ability to answer for a decision after the people who made it have moved on.
Free download
Get the full paper
Read The Responsible Performance System in full, with every claim carrying an evidence grade and the full reference list. Give your name and email and the PDF opens straight away.
Frequently asked questions
Does AI actually predict sports injuries?
Not reliably at the individual level on the published evidence. Hamstring injury models tested across seasons returned a median area under the curve of 0.52, where 0.5 is equivalent to a coin toss. The acute to chronic workload ratio adds nothing to an intercept-only model even within its own training sample. Load records remain useful for describing what an athlete did. Load ratios as individual injury predictors do not currently support the claims made for them.
Is sports AI regulated?
Barely, and that is the difficulty. The word athlete does not appear in the European Union’s Artificial Intelligence Act, and sport appears only in its recitals. At the time of writing no governing body had published a binding instrument on performance AI. General data protection law does apply, and it already grants athletes access, portability, rectification, erasure and human intervention where automated decisions are made about them.
When does a sports AI tool become a medical device?
The line falls at automated reasoning about a person’s future health. A dashboard that reports what an athlete did sits outside the medical device regime. A tool that applies automated reasoning to output an individual’s future risk of injury has described itself into the regulatory definition, which names prediction, prognosis and injury expressly. This is a question for qualified regulatory advisers rather than a matter to settle in a procurement meeting.
Is a human sign-off enough oversight for automated decisions?
No. Automation bias is well documented in clinical decision support, and the conditions that worsen it are workload, task complexity and time pressure, which describe a training ground precisely. A signature is a record, not a safeguard. Oversight requires that the person can see the reasoning, has a real alternative available, and has the time and standing to disagree.
What should a club measure instead of a readiness score?
Completed planned training weeks. Athletes who completed more than 80 per cent of planned weeks were seven times more likely to hit a performance goal, at an area under the curve of 0.72. It is countable, auditable, and it outperforms proprietary composite scores on published numbers.
About this series
This is the first of seven papers examining how artificial intelligence can be used responsibly in performance sport. It supplies the assurance framework that the six sport-specific papers draw on. The others cover elite track and field, grassroots athletics, grassroots and academy football, triathlon and endurance, elite football and golf.
Legal and regulatory notice. This article and the paper it summarises describe legislation, regulatory guidance and governing-body requirements. Nothing here constitutes legal advice, and the provision of legal advice sits outside the terms of any engagement with the author or with Athleet.AI. The material is presented to support discussion and further review by qualified advisers. Every club, athlete and scenario used as an illustration is synthetic.