The first Games to use large language models. AI-traced curling stones, 360-degree replays across 17 sports, and automated highlights in minutes. It also monitored abuse aimed at athletes, and caught only a fraction of it.
Published: Saturday, 1 August 2026 | BrandClickX News Desk
Note: The Milano Cortina 2026 Winter Olympics ran from 6 to 22 February 2026. This is an assessment of technology that has already been deployed, not a preview.
Summary
The Milano Cortina 2026 Winter Olympics marked the first use of large language model technology at an Olympic Games. Alibaba Cloud, working with Olympic Broadcasting Services and the IOC, deployed AI across 17 sports for replays, highlight generation and computer-vision graphics. OMEGA used AI computer vision for timing data across all 116 events. AI also supported judging trials, cybersecurity and online abuse monitoring with mixed results.
What Was Actually Deployed
Six distinct systems, across broadcasting, timing, judging, fan services, security and athlete protection.
| Area | What ran |
| Broadcasting | 360-degree replays, Spacetime Slices, automated highlights, AI tracking cameras |
| Timing and data | OMEGA Computer Vision, athlete-worn sensors, image tracking |
| Judging | AI-assisted rotation counting in figure skating; jump measurement in aerial events |
| Fan services | “Olympic GPT” on Olympics.com; Intelligent Pin Trading Station in the Village |
| Security | AI-supported threat monitoring by Italy’s National Cybersecurity Agency |
| Athlete protection | IOC AI system flagging abusive social media content |
IOC chief technology and information officer Ilario Corna framed it as a threshold moment: Milano Cortina 2026 marked a defining point in AI’s integration into the Olympic Movement, including the first use of LLM technologies at the Olympics.
Broadcasting: Where the Money Went
Olympic Broadcasting Services deployed AI across 17 sports and disciplines, including ice hockey, freestyle skiing, figure skating and ski jumping.
The specific systems:
- 360-degree replays with stroboscopic analysis, building on the BulletTime effects introduced at Beijing 2022
- Spacetime Slices, a new capability layered on top of frame-freeze and slow-motion views
- Automated highlights generated within minutes of key action, delivered ready-to-publish to every platform
- AI tracking cameras delivering individual athlete coverage to rights holders
- Computer-vision graphics converting performance into data overlays
- Audio enhancement isolating signature sounds while preserving ambient crowd noise
The curling application is the clearest example of why this matters. AI traced the path of each rock with precision, letting viewers follow tactics that were previously legible only to people who already understood the sport.
OBS chief executive Yiannis Exarchos made the reasoning explicit: winter programmes contain many sports that are popular in some regions and barely known in others. The technology is a translation layer for unfamiliar disciplines.
His colleague’s framing is the one worth keeping: technology is not the goal, it is the enabler.
Timing: 90 Years and 40,000 Images a Second
OMEGA provided timing for all 116 events across 16 sports, including ski mountaineering for the first time.
Milano Cortina marked 90 years of OMEGA timekeeping at the Winter Games.
The equipment list is the least speculative part of this story because it is measurement rather than inference:
- Scan’O’Vision ULTIMATE photofinish cameras capturing up to 40,000 digital images per second at the finish line
- Electronic Starting Pistol and Snowgate Technology
- Computer Vision capturing continuous performance data live bobsleigh speed, live positions in speed skating
- Athlete-worn motion sensors, built on systems first deployed in 2018
New systems debuted in figure skating, bobsleigh, ski jumping and big air.
The distinction between this and generative AI matters. OMEGA’s computer vision extracts data from what physically happened. It does not produce content.
Judging: The Contested Application
The IOC used AI to support figure skating judging helping identify the number of rotations completed in a jump with extension to big air, halfpipe and ski jumping.
In those aerial disciplines, automated systems measured jump height and take-off angles.
This is where the technology moves from describing sport to influencing outcomes, and reaction was divided.
The case for it: human judging panels in figure skating and gymnastics have documented problems with bias, inconsistency and transparency. Counting rotations at speed is precisely the kind of task machines do more reliably than people.
The case against it, raised at the first Olympic AI Forum in November 2025: an AI system trained on existing elements does not know what to do when an athlete introduces a genuinely new trick. Novelty is often exactly what judged sports are supposed to reward.
That objection has not been resolved. It is a structural limitation rather than a bug to be patched.
The Honest Failure: Online Abuse
The IOC ran an AI system to monitor abusive content aimed at athletes. Japan’s delegation logged around 62,000 abusive posts anyway.
The Japanese Olympic Committee recorded that figure from 18 January onwards. Of those, 1,055 prompted formal takedown requests, and 198 were removed.
That is a removal rate of roughly 0.3% of flagged abuse.
Japan had its best Winter Games since Nagano 1998 20 medals including four golds and delegation chief Hidehito Ito still spent a press conference in Milan asking people to stop, saying that directing hurtful words at athletes violates their dignity and undermines their performance.
Japan deployed its own staff in Italy, monitoring platforms around the clock and using AI to flag content for removal.
This is the most instructive data point in the entire AI story at these Games. Detection worked. Enforcement did not. The bottleneck was not identifying abuse it was platforms acting on it.
Security: AI on Both Sides
Italy’s National Cybersecurity Agency prepared specifically for attackers using AI to scale cyber operations.
The ACN, created in 2021, faced its first major global event. Its specialists spent more than a year tracking criminal forums, social channels and underground markets to anticipate threats.
The stakes, per ACN director of cyber operations Gianluca Galasso: around three billion viewers and 1.5 million ticketed spectators.
Identified risk areas were streaming disruption, ticketing systems and event websites.
It was not theoretical. Italian foreign minister Antonio Tajani announced that several cyber-attacks targeting ministry pages and Olympic-linked sites, including hotels in Cortina d’Ampezzo, had been neutralised. He attributed them to Russian hackers, and a pro-Russian group claimed responsibility on Telegram in retaliation for Italy’s support for Ukraine though authorship could not be independently verified.
The IOC declined to comment, with communications director Mark Adams citing standard practice on security matters.
The Fan-Facing Layer
“Olympic GPT” answered questions, interpreted rules and summarised content on Olympics.com.
Reporting emphasised that it was built on verified Olympic data, with the priority being accuracy and reliability rather than open-ended generative output. That is a narrower deployment than the branding suggests, and deliberately so.
More novel was the Intelligent Pin Trading Station in the Olympic Village, powered by Alibaba’s Qwen large language model, which let athletes trade pins through a shared pool.
Trivial on its own. Strategically notable as an example of LLMs used in a physical on-site experience rather than on a website.
The Framework Behind It
The IOC published its Olympic AI Agenda in April 2024, and held the first Olympic AI Forum in November 2025.
The Agenda commits to AI usage that supports equity and sustainability, remains reliable, operates transparently, and aligns data and privacy practices with public expectations including limiting what is collected.
Corna has described convening a working group months in advance specifically to ask those questions before deploying anything.
The counterweight comes from academic work on Paris 2024, which examined AI surveillance at that Games and argued that sport mega-events function as testing sites for security technologies. That research explicitly flagged Milano Cortina 2026 and Los Angeles 2028 as the next decision points, noting that whether such systems are adopted or rejected is itself significant.
Milano Cortina’s AI security deployment was framed as defensive protecting infrastructure against attackers. That is a different proposition from crowd surveillance, and the distinction is worth preserving rather than collapsing.
Timeline
| Date | Development |
| April 2024 | IOC publishes the Olympic AI Agenda |
| Nov 2025 | First Olympic AI Forum held |
| 6 Feb 2026 | Games open; ACN reports neutralised cyber-attacks |
| 6–22 Feb 2026 | AI deployed across broadcasting, timing, judging and fan services |
| 18 Feb 2026 | Japan reports 62,000 abusive posts despite AI monitoring |
| 22 Feb 2026 | Games close |
| 2028 / 2030 | Los Angeles and French Alps are the next deployment points |
Expert Analysis
Three conclusions the evidence supports.
Broadcasting is where AI genuinely delivered. Automated highlights in minutes, 360-degree replays across 17 sports, and rock-tracing in curling are concrete improvements to how audiences understand sport. Exarchos’s point about unfamiliar winter disciplines is the strongest argument in the whole AI-in-sport debate this technology makes niche sports legible to people who have never watched them.
Judging remains genuinely unresolved. Counting rotations is a solvable problem. Evaluating a trick nobody has performed before is not, and no amount of additional training data fixes it, because the definition of the problem is novelty. The IOC has proceeded carefully here, using AI to support judges rather than replace them, which is the right posture for an unsolved problem.
The abuse figures are the story nobody is telling. An AI system flagged tens of thousands of abusive posts, and 198 came down. That gap is not a failure of detection. It is a failure of what happens after detection and it demonstrates that AI capability without enforcement authority produces documentation rather than protection.
That pattern will repeat wherever AI is deployed against a problem it can identify but not resolve.
Key Takeaways
- Milano Cortina 2026 was the first Olympics to use large language model technology
- AI broadcasting ran across 17 sports including ice hockey, figure skating and ski jumping
- Automated highlights were generated within minutes of live action
- OMEGA provided AI-assisted timing for all 116 events across 16 sports
- Photofinish cameras captured up to 40,000 images per second
- AI supported figure skating judging by counting jump rotations
- Japan logged 62,000 abusive posts; 198 were removed
- Italy’s ACN prepared for AI-assisted cyberattacks; several were neutralised
- Los Angeles 2028 and the French Alps 2030 are the next deployments
Frequently Asked Questions
How was AI used at the 2026 Winter Olympics?
Across broadcasting, timing, judging support, fan services, security and athlete protection. Alibaba Cloud and Olympic Broadcasting Services deployed AI in 17 sports for replays, automated highlights and computer-vision graphics.
Was this the first Olympics to use AI?
No, but it was the first to use large language model technology. AI had been deployed at Tokyo 2020, Beijing 2022 and Paris 2024, with Milano Cortina marking a step change in scale and in LLM adoption.
Did AI judge Olympic events?
Not independently. AI supported human judges, helping identify the number of rotations in figure skating jumps and measuring height and take-off angles in big air, halfpipe and ski jumping.
What is Olympic GPT?
An AI service on Olympics.com that answered questions, interpreted rules and summarised content, built on verified Olympic data. Reporting emphasised accuracy over open-ended generative output.
How did AI change Olympic broadcasting?
Automated highlights were produced within minutes, 360-degree replays with stroboscopic analysis ran across 17 sports, AI tracking cameras followed individual athletes, and computer-vision graphics turned performance into data overlays.
Did AI stop online abuse of athletes?
No. The Japanese Olympic Committee recorded around 62,000 abusive posts from 18 January, despite IOC AI monitoring. Of those, 1,055 prompted takedown requests and 198 were actually removed.
What role did AI play in Olympic timing?
OMEGA used Computer Vision, image tracking and athlete-worn sensors to capture live data such as bobsleigh speed and speed skating positions, alongside photofinish cameras capturing 40,000 images per second.
What is the Olympic AI Agenda?
An IOC framework published in April 2024 setting out how AI should be used across the Olympic Movement, committing to equity, sustainability, reliability, transparency and limits on data collection.
Conclusion
The most useful way to read AI at Milano Cortina is by separating what it measured from what it decided.
Measurement worked. Timing, tracking, replays, highlights, rock traces in curling all of it made the Games more legible, particularly for sports most viewers see once every four years.
Decision-making is where it gets harder. Judging support is carefully scoped and still unresolved on novelty. And the abuse-monitoring figures show a system that identified a problem at scale and removed 0.3% of it.
Los Angeles 2028 and the French Alps 2030 will be built on these foundations. The broadcasting case is proven. The rest is still an argument.
