AI cameras at events: what the EU AI Act allows

Cameras that analyse images rather than simply delivering them have arrived at events: they count visitors, measure density in front of the stage or report when someone enters a closed-off area. With the EU AI Act, organisers and operators face a new question: is this still allowed – and if so, what obligations come with it?

This article explains what the AI Act means for video analytics at events and on sites, and what the General Data Protection Regulation (GDPR) requires in addition. It reflects the position on 28 October 2026 and is not legal advice.

How the AI Act is structured

Regulation (EU) 2024/1689 regulates AI according to risk. Some uses are prohibited, others are classed as high-risk and subject to strict obligations, some carry transparency obligations, and the rest face hardly any specific requirements.

Two roles matter. The provider develops an AI system and places it on the market. The deployer uses it under its own authority in a professional context (Article 3(3) and (4)) – at an event, usually the organiser or the security contractor.

The obligations apply in stages. In July 2026, amending Regulation (EU) 2026/1744 (the “Digital Omnibus”) postponed the deadlines for high-risk systems:

Since / from	What applies
2 February 2025	Prohibitions under Article 5, AI literacy under Article 4
2 August 2026	Transparency obligations under Article 50 and further parts of the Regulation
2 December 2027	Obligations for high-risk systems under Annex III (previously 2 August 2026)
2 August 2028	Obligations for high-risk systems in regulated products under Annex I

Breaches of the prohibitions can be fined up to EUR 35 million or 7 % of worldwide annual turnover, whichever is higher (Article 99(3)).

What is prohibited

For cameras at events, three prohibitions in Article 5(1) are particularly relevant:

• Real-time remote biometric identification in publicly accessible spaces for law enforcement – with narrow exceptions, such as the targeted search for missing persons or preventing a terrorist attack (point (h)).
• Emotion recognition in the workplace and in education, except for medical or safety reasons (point (f)). At events this concerns, for example, stewards, stand staff and the organiser’s employees.
• Biometric categorisation that infers characteristics such as religion, political opinions or sexual orientation from biometric data (point (g)).

There are also general prohibitions, for example on manipulative techniques or on scoring people by their social behaviour.

What counts as high-risk

Annex III lists the areas in which AI systems are classed as high-risk. Two points matter for video analytics:

• Point 1 – biometrics: remote biometric identification (not mere verification that someone is who they claim to be), biometric categorisation by sensitive characteristics and emotion recognition, where these are not already prohibited.
• Point 2 – critical infrastructure: AI used as a safety component in the management and operation of critical digital infrastructure, road traffic or the supply of water, gas, heating and electricity.

From December 2027, high-risk systems face extensive obligations: risk management, technical documentation, human oversight, logging and, for deployers, use in line with the instructions, among others. Operators of critical facilities should therefore check carefully whether their video analytics serve as a safety component within the meaning of point 2.

Counting is not identifying

The key lies in the definitions. Under Article 3(35), biometric identification is the automated recognition of human features in order to establish a person’s identity by comparing them with stored data. A remote biometric identification system matches people against a reference database without their active involvement (Article 3(41)).

Typical safety functions at events do none of this:

Function	Does it establish who someone is?
Measuring persons per m² in an area	No – it counts people
Counting arrivals and departures at an entrance	No – it counts movements
Reporting a person in a restricted area	No – it reports that someone is there
Reporting a bag with nobody nearby	No – it reports an object

In our assessment, such functions therefore fall neither under the biometric prohibitions nor under Annex III point 1. What counts, however, is always the specific purpose. Anyone who wants to analyse the behaviour or state of individuals needs to look more closely – for example, whether emotions or intentions are being inferred from biometric data (Article 3(39)).

What still applies: data protection and AI literacy

A function falling outside the prohibitions does not mean that no rules apply. Camera images show people who may be recognisable. That makes them personal data, even if nobody is identified (Article 4(1) GDPR). It follows that:

• Legal basis: private operators usually rely on their legitimate interest in safety (Article 6(1)(f) GDPR) and must balance it against visitors’ interests.
• Data minimisation: capture only what the purpose requires and keep it no longer than necessary (Article 5) – and build this into the technology from the start (Article 25).
• Notices: signs before the monitored area naming the controller and purpose (Article 13); the guidelines of the European Data Protection Board describe what they should look like.
• Data protection impact assessment: mandatory for systematic monitoring of publicly accessible areas on a large scale (Article 35(3)(c)).

On top of this comes AI literacy under Article 4 of the AI Act. Since the 2026 amendment, providers and deployers must take measures to support the AI literacy of their staff. In practice, anyone in the control room who sees AI warnings should know what the system detects, where its limits are and who decides.

Questions to ask a provider

• Does the system establish the identity of people or match them against databases?
• Does it infer emotions, intentions or sensitive characteristics from images?
• Where does the analysis run – on site or in a cloud, and in which country?
• Is video stored? If so, for how long and who has access?
• Can parts of the image be masked, such as the windows of homes?
• Does a person decide when a warning appears, or does the system act on its own?
• Does the provider supply material for a data protection impact assessment and for training?

How MOVETOS does it

In the MOVETOS OPS platform, the AI runs on hardware on site – on your server or in the mobile control room; no cloud is required. Our own AI model counts heads in the camera image and calculates crowd density from the measured ground area. The detections in live operation are crowd density, person in a restricted area and unattended luggage.

No video is recorded. What is stored are figures and, only when there is a warning, an evidence photo. Parts of the image that must not be observed can be masked as privacy zones. The system suggests, a person approves: every warning is assessed by someone in the control room, and the operation is fully documented. How counting works in detail is explained in Counting visitors with cameras.

Checklist for organisers and operators

• Define the purpose: what should the video analytics do – count, measure density, monitor restricted areas?
• Classify: is anyone identified, categorised or assessed for emotions? If not, record in writing why not.
• Critical facility? Check whether the analysis serves as a safety component under Annex III point 2.
• Document the legal basis and the balancing of interests.
• Carry out a data protection impact assessment where publicly accessible areas are monitored on a large scale.
• Put up notices and set up privacy zones.
• Limit storage: no video unless it is needed; set retention periods for evidence photos.
• Train staff: what does the AI detect, where are its limits, who decides?

Conclusion

The AI Act is aimed above all at systems that identify or categorise people or analyse their emotions. In our assessment, counting people, measuring density and monitoring restricted areas are not among them. Data protection, clear purposes and trained staff remain mandatory all the same. Organisers who classify their systems early, document their reasoning and choose data-minimising technology can use AI at events – and justify it later.