IOC Trustworthy AI Framework: What It Does—and Does Not Prove

AI can analyse an athlete or help explain a decision—but who checks how it uses data and whether its output is fair? The IOC’s new Trustworthy AI Framework sets out seven principles and a practical checklist for responsible AI across the Olympic Movement.

The important starting point is modest but useful: this is published governance guidance, not evidence that a particular AI product works well in every sport. The official 22-page Trustworthy AI Framework for the Olympic Movement, dated 28 September 2026, is accompanied by training materials and a checklist. Together, they give organisations a shared way to question an AI system before and during use. [1] [2]

Is the IOC Trustworthy AI Framework a published document?

Yes. The IOC lists the framework, related training materials and an Excel checklist as published resources dated 28 September 2026. The document’s full name is Trustworthy AI Framework for the Olympic Movement. It is intended to support responsible choices about AI in sport across the Olympic Movement, rather than to describe a single product or competition system. [1] [2]

That matters because a framework is a tool for governing work, not a performance result. It can help a federation, club, event organiser or other organisation ask sharper questions about an AI supplier, data flow or decision process. It does not by itself verify that a particular system has been independently tested, is appropriate for a particular athlete group or will deliver a sporting outcome.

Illustrative photo: All-weather running track at Dalin Sports Park, Chiayi, Taiwan, photographed in 2017. This WebP adaptation was resized and converted from the original; no crop was made. Photo: Malcolm Koo (Mk2010), CC BY-SA 4.0, via Wikimedia Commons. This is not an IOC, Olympic-event, framework-launch or AI-deployment photograph. [5] For a broader explanation of how models, data and human judgement fit together, see AI in Sports: Uses, Examples, Benefits and Limits. This page stays narrower: it explains what the IOC document asks organisations to govern.

How is it different from the Olympic AI Agenda?

The 2024 Agenda set strategic direction; the 2026 framework turns part of that direction into a practical governance aid. They are connected, but they are not two names for the same publication.

The Olympic AI Agenda is a 10-page IOC strategy document dated 11 April 2024 and launched on 19 April 2024. It describes ambitions, guiding principles and potential high-impact areas for Olympic AI, including training, officiating, safeguarding and broadcasting. Those are areas of interest and direction, not measured outcomes promised by the later framework. [3] [4]

If your question is about how an organisation should document authority, challenge an output or review a supplier’s claims, the framework is the more direct reference. If your question is how an analytics model can turn sports data into a useful decision, the separate Sports Analytics guide provides that wider operating context.

What are the seven Trustworthy AI Principles?

The principles are a set of distinct control questions, not a single score. In practice, a system can be clear to users yet still be unreliable; it can be technically secure yet still unfair to a group. Treat each principle as a separate line of review. [1]

1. Prioritise safe and secure AI for sport. Start with plausible harms and security weaknesses. Ask what could go wrong if a system is misused, accessed without permission or relied on outside its intended purpose.
2. Protect personal data with respect and care. Identify what athlete, staff or other personal data enter the tool, who may access them, why the data are needed and how long they will be kept. This is especially relevant when performance-tracking or wearable data are involved; our guide to wearable technology in sports explains why a measurement is not the same as a decision.
3. Deliver robust and reliable AI for sport. Check whether the system is dependable for the stated task and conditions. A tool trained on one sport, level of competition, camera setup or population may not be equally dependable in another setting. Record what it was tested against and what it cannot reliably handle.
4. Make the use of AI clear and transparent. People affected by an AI-supported process need a clear explanation of its role, its inputs and the limits of its output. “AI-assisted” is not enough information for an athlete or decision-maker to understand what has happened.
5. Govern AI with accountability and responsibility. Name the organisation and people who can approve use, interpret an output, respond to a concern and stop the tool. The framework is explicit that responsibility remains with organisations developing, procuring or using AI. [1] [2] For a decision-focused extension of this idea, see responsible AI controls for player decision support.
6. Ensure AI is fair and equitable for all. Examine whether data, design choices or deployment conditions could disadvantage people or groups. Fairness is not established because a dashboard looks neutral. It needs a defined question, appropriate evaluation and a route to investigate a concern.
7. Support sustainable AI for the future of sport. Consider the ongoing resources and consequences of an AI system, rather than treating sustainability as an afterthought once it is in use.

These principles do not replace sport-specific expertise. They give that expertise a better structure: define the use, test its limits, explain it, assign responsibility and revisit it when conditions change.

How does the IOC checklist turn principles into questions?

The checklist moves from risk to review to recommendations. The framework describes three steps: evaluate an AI system’s inherent risk, apply the seven principles across its lifecycle, then receive recommendations based on the answers. [1]

That makes the checklist useful at more than one point. Before procurement, it can help a buyer ask what evidence a supplier should provide. Before deployment, it can prompt a team to clarify data access, human authority and fallback arrangements. During use, it can make monitoring and correction part of the process rather than an afterthought.

But a checklist result should not be mistaken for a seal of approval. The IOC describes it as a self-assessment aid; it is not an independent audit, a certification or a substitute for legal, technical, clinical or safeguarding review where those are needed. Its value depends on candid inputs, evidence behind the answers and the willingness to act on gaps.

For example, an organisation considering a generated training schedule could use the checklist to ask who reviews the recommendation, which data it uses and how a participant can raise a concern. It should still assess the proposed output in context. AI-generated training plans need human review rather than automatic adoption.

Who is responsible when an Olympic Movement organisation uses AI?

Responsibility stays with the organisation, not with the framework or the software. The IOC says organisations that develop, procure or use AI remain responsible for safe and ethical deployment. [1] [2]

A practical owner should therefore be able to answer a few plain questions: What decision is this tool meant to support? Who is allowed to use it? Which data are authorised for that use? Who checks the output before it affects a person? What documentation supports the system’s limitations? Who can correct a record, challenge a result or pause the tool?

This also means setting a stop or fallback condition. If data quality changes, a model performs unexpectedly or an output cannot be explained well enough for the decision at hand, the organisation needs a defined alternative. Athlete performance tracking is useful here: it frames measurements as inputs to an accountable decision, not as a verdict on their own.

What the framework can and cannot establish about athlete safety

The framework can prompt a structured review of potential harms; it cannot prove prevention or clinical validity. It discusses risks, including possible physical and psychological harms, and asks users to assess an AI system’s risk. That is appropriate governance work, but it is not outcome evidence. [1] [2]

The framework does not validate an injury-prediction model, diagnose an injury, show that an AI system predicts injuries reliably in every setting or prove that AI prevents injuries. Nor does it establish that a tool improves athlete safety or performance. Those claims need evidence about the named system, population, use case and outcome.

That distinction is more than careful wording. A prediction can be statistically associated with a future event without showing that acting on it changes that event. For a worked evidence boundary, see Injury-Risk AI Study Separates Prediction From Causal Claims. The framework can help an organisation ask whether a tool deserves further scrutiny; it is not the scrutiny itself.

What to check before adopting an AI tool in sport

Use the framework as a decision checklist, not as a shortcut to approval. The following reader-facing questions translate its principles into a procurement or operating conversation. They are editorial prompts based on the IOC framework, not an IOC certification process. [1]

• Purpose: What defined decision will the tool support, and what must it never decide on its own?
• People and data: Whose data are involved, what is the permitted basis for use, and who can access or export them?
• Risk and limits: What harms, errors or misuse cases are plausible? Which settings or groups were not adequately evaluated?
• Evidence: What testing, documentation and independent assessment can the supplier provide for this exact use?
• Transparency: Can an affected athlete, staff member or official understand the tool’s role and question a result?
• Accountability: Who owns the decision, who reviews exceptions and who responds when something goes wrong?
• Monitoring and fallback: How will performance and data quality be reviewed over time, and when will the system be paused or replaced by a non-AI process?

Keep the answers with the procurement record or operating procedure, not only in a presentation. A pilot can be useful when it is designed to resolve a decision uncertainty; our guide to turning a SportsTech pilot into a procurement decision explains how review gates and evidence can support that work.

Where the framework fits in the wider AI-in-sports discussion

This framework is a governance reference inside a much bigger AI-in-sports conversation. It does not replace technical evaluation, sport-specific policy, data-protection duties or human judgement. It gives organisations a common structure for asking whether an AI system is suitable, understandable and accountable before its output carries too much weight.

Read the 2024 Agenda for the IOC’s strategic context and the 2026 framework for its seven principles and checklist. Then apply the questions to the actual tool, data and decision in front of you. That is more useful than treating a framework as proof that any AI system is ready, fair or safe. [1] [3]