Sunday, 20 September 2026
The International Olympic Committee is using artificial intelligence across athlete protection, talent identification, event operations and fan engagement, but its broader strategy offers a lesson extending well beyond sport: organisations trying to scale AI need to decide where the technology creates value before deciding which technology to deploy. Speaking at Ai4 2026 in Las Vegas, Alejandro Merino-Madrid, Head of AI at the International Olympic Committee, described how the organisation has approached AI transformation through a deliberately restricted set of priorities rather than encouraging experimentation across every possible function.
The IOC’s strategy centres on supporting athletes and competition, making AI capabilities more widely accessible within sport, improving the organisation and sustainability of the Olympic Games, enhancing the fan experience and increasing the organisation’s own efficiency. The principle is straightforward. AI projects falling outside those areas do not automatically receive resources simply because the technology is available. In a corporate environment increasingly surrounded by generative AI pilots and agentic experiments, that discipline may become increasingly important because organisations can now build prototypes relatively easily, while determining whether those prototypes support strategic objectives is considerably harder.
Athlete protection provides one of the clearest examples of the IOC’s approach. At the Olympic Games Paris 2024, an AI-assisted cyber-abuse protection service monitored millions of social-media posts and comments directed at athletes and officials. The system analysed approximately 2.4 million posts and comments, flagged more than 152,000 as potentially abusive and ultimately identified more than 10,200 that were verified as abusive. Some 353 athletes and officials were directly targeted and offered safeguarding and mental-health support.
Rather than relying on athletes or their families to identify abusive content after seeing it, the system allowed potentially harmful material to be detected and escalated at scale. For the IOC, the objective was therefore not simply content moderation but protecting athletes’ ability to concentrate on competition. The programme demonstrates an important difference between using AI because it is technologically interesting and using it to address an identifiable operational problem. The IOC first defined the outcome it wanted — reducing athletes’ exposure to online abuse — and then applied AI because the volume and speed of social-media activity made purely manual monitoring impractical.
A second initiative illustrates how AI could address inequality in access to sporting opportunities. The IOC has tested AI-assisted talent identification in Senegal, where young participants performed predefined physical exercises while computer-vision systems analysed performance indicators. The purpose was to identify potentially promising athletes in locations where conventional scouting networks may have limited reach. The concept addresses a structural problem within sport because talent does not necessarily emerge in areas containing sophisticated training centres, professional coaches or well-funded scouting systems.
AI potentially allows basic screening to be conducted using relatively inexpensive equipment before specialised human coaches become involved. Merino-Madrid argued that this could reduce some of the geographical disadvantages facing young athletes in remote or underserved areas, while the same principle may apply even in wealthy countries containing large rural populations where conventional scouting is expensive. The broader business analogy is significant because many organisations have valuable information, customers, employees or opportunities distributed across locations that cannot economically receive the same level of human attention. AI may allow companies to widen the area in which they search for opportunities while reserving expensive specialist resources for cases identified as particularly promising.
Another application involves real-time analysis of competition. The IOC and its technology partners are increasingly using computer vision and data processing to convert events on the field of play into information that can be made available to broadcasters, athletes, officials and fans almost immediately. This can include tracking movement, measuring performance and creating digital representations of athletes during competition. Such information can help broadcasters explain why a particular performance was exceptional, provide athletes with additional feedback and support officials in sports where movements occur too quickly or precisely for the unaided human eye to assess consistently.
The ambition is not simply to create more statistics. Data that previously existed invisibly inside a competition can be converted into narratives that help viewers understand what they are watching. This reflects another broader change occurring across the economy. Organisations have spent years accumulating enormous datasets, frequently displaying them through dashboards. AI increasingly allows those datasets to become conversational and contextual, bringing relevant information directly to the person making a decision rather than expecting that person to search for it manually.
However, the IOC’s presentation placed as much emphasis on trust as on capability. Merino-Madrid argued that AI is moving closer to areas where incorrect decisions carry meaningful consequences. Systems may influence athlete training, assist competition officials, analyse personal information or shape what millions of spectators see. The closer AI moves to these functions, the more important it becomes to determine what the system is allowed to do and how its decisions can be assessed.
The IOC has therefore developed a Trustworthy AI Framework intended to guide organisations within the Olympic Movement when evaluating and deploying AI. The framework combines education, governance principles and practical checks intended to help organisations ask questions about areas such as data privacy, security and appropriate use before adopting a system. This is particularly relevant because the Olympic Movement includes organisations with dramatically different technological capabilities. A large international federation may employ extensive technology and legal teams, while a smaller National Olympic Committee may have far fewer resources available to assess a proposal from a major technology provider.
Without a common framework, the two organisations could reach very different conclusions about the same technology simply because one has more expertise available. The risk runs in both directions. An organisation may reject useful technology because it cannot assess the risks, or it may adopt a system it does not fully understand and subsequently discover that the technology creates privacy, security or governance problems. The IOC’s response is therefore not to attempt to eliminate risk entirely but to give decision-makers a repeatable method for evaluating it.
That philosophy translates directly into enterprise AI. Many companies currently face a fragmented technology market involving large language models, AI agents, cloud platforms and specialist applications. Procurement teams and business units are being approached with solutions promising dramatic productivity gains, often before internal governance structures have developed sufficiently to evaluate them. A clear checklist covering who owns the data, where it is stored, who can access it, what decisions the AI can make and how the system can be audited may therefore be more valuable than another experimental AI application.
Data readiness is another obstacle familiar well beyond the IOC. Merino-Madrid said the organisation discovered that possessing large quantities of information did not mean the information was ready for AI. Data may be incomplete, inconsistently structured, duplicated or outdated. Attempting to scale AI on top of such information can simply reproduce those weaknesses at greater speed. The lesson, in his view, is that organisations should invest in their information foundations before attempting widespread deployment.
Cleaning, structuring and governing data rarely attracts the attention given to a new AI interface, but it frequently determines whether that interface can eventually operate reliably. The same applies to security. Agentic systems can potentially access information and perform actions on behalf of users, making role-based access increasingly important. An AI assistant used by a senior executive and a junior employee cannot necessarily have access to the same underlying information merely because both interact with the same interface.
Permissions therefore need to remain attached to the person making the request rather than simply to the AI system processing it. This is becoming one of the most important challenges surrounding enterprise agents. Traditional software usually gives individuals explicit access to particular applications or databases. An agent potentially sits across several systems simultaneously, meaning organisations must ensure that convenience does not allow employees to bypass existing information controls.
Regulatory complexity adds another layer. The IOC operates internationally across jurisdictions where rules affecting privacy, data processing and artificial intelligence differ and continue to evolve. Companies operating across multiple markets face the same problem. A system that is acceptable in one country may require different controls in another, while the regulatory position itself can change faster than the technology procurement cycle. Governance therefore cannot be treated as a document created once at the beginning of an AI programme. It has to evolve as regulation, technology and organisational use change.
Merino-Madrid summarised the IOC’s transformation approach around three basic ideas: know where the organisation is trying to go, begin delivering tangible value and establish guardrails that prevent innovation from damaging trust. The first element is strategic focus. An organisation should determine which business problems matter before deciding where AI should be used. Without that filter, resources can easily disappear into hundreds of demonstrations that produce little lasting operational value.
The second is implementation. An AI strategy that remains confined to presentations and proof-of-concept projects does not transform an organisation. Companies need applications capable of producing measurable operational improvements and then need to learn from deploying them in real working environments. The third is trust. Once AI becomes involved in decisions affecting employees, customers, financial outcomes or sensitive information, governance becomes part of the product itself rather than a separate compliance exercise.
For companies following the rapid development of generative and agentic AI, the IOC’s experience offers a useful counterweight to the prevailing pressure to move as quickly as possible. Speed matters, but speed without direction can simply increase experimentation costs. AI can only scale sustainably if organisations know what they are trying to improve, have information reliable enough to support the system and understand the boundaries within which the technology is permitted to operate.
The Olympic example is particularly instructive because the organisation has to reconcile stakeholders ranging from elite athletes and broadcasters to international federations, technology companies and more than 200 National Olympic Committees. Those participants differ substantially in resources, technical expertise and willingness to adopt AI. That complexity can make transformation slower, but it can also create stronger systems if solutions are designed to work across different levels of technological maturity.
The lesson for business is that the most sophisticated AI architecture is not necessarily the most valuable one. Sometimes a governance checklist, clearly defined permissions and properly maintained data can matter more than the newest model. The IOC’s AI programme ultimately illustrates an emerging reality of enterprise transformation: AI technology is becoming easier to obtain, while organisational clarity, trusted information and responsible implementation are not. As artificial intelligence moves from isolated tools towards systems capable of influencing operational decisions, those less visible capabilities may become the real competitive advantage.
Source: CIJ.World Research & Analysis Team
