Spain notifies EU of DGOJ risk-detection algorithm under urgent procedure
Regulation3 min read
Summary
Spain notified the European Commission on 17 November 2026 of a draft DGOJ resolution establishing a common system to identify gambling-related risk. The proposed system would be mandatory for operators supervised by the DGOJ, covering all gambling products they offer, and would replace a draft notified in 2025. Spain requested the EU’s urgent procedure, citing a review of commercial communications rules and increased self-exclusion registrations.
Spain's government has notified the European Commission of a draft DGOJ resolution establishing a system to detect gambling-related risk. The notification was registered on 17 November 2026 under reference 2026/0654/ES. Spain requested the urgent procedure under Directive 2015/1535, meaning the ordinary standstill period will not apply.
The proposed system would be mandatory for DGOJ-supervised operators across all gambling products they offer. Operators could retain their own protocols alongside it, but not in place of the common system. An accompanying memorandum says the aim is to replace companies' existing systems with uniform criteria and provide the same level of protection across the market.
The DGOJ previously notified the text on 12 December 2025 under reference 2025/0747/ES. It is presenting a new procedure following comments received and substantial changes to the wording. To justify urgency, the regulator cited a review of Royal Decree 958/2020 on commercial communications and an increase in self-exclusion registrations. The mechanism develops Royal Decree 176/2023 on safer gambling, whose article 24 requires operators to detect risk behaviour and permits the regulator to set criteria by resolution.
The memorandum cites data from 55 active operators, of which 36 identify fewer than 1% of accounts as at risk and six identify more than 4%. The proportions range from 0% to 13.6%.
The system would examine a period of 183 calendar days ending on the final day of each month. Operators would run it on all player accounts during the first seven days of each month. The first step would exclude occasional players, including those with fewer than two deposits or fewer than two days of activity, and those who play very little. A direct detection rule would then be applied, followed by an XGBoost machine-learning model supplied by the DGOJ for the remaining players, with a common probability threshold for all operators.
Players classified as at risk would be subject to the measures in articles 26 to 30 of Royal Decree 176/2023 for at least six months. The measures would be lifted after that period if the next application of the model no longer detected risk, or extended for at least another six months if it did.
The classification would not be a clinical diagnosis and could not alone justify other decisions about a player. An account suspension would require a specific, reasoned assessment. Operators would have to tell players the date and period examined, the main factors behind their classification in understandable language, the applicable measures and the conditions for lifting them. Players could request an operator review, which would have to be resolved within one month, and could complain to the DGOJ if they disagreed with the outcome. Neither a review nor a complaint would suspend the measures. The data and classification could be used only for protection, not for commercial, promotional, advertising or segmentation purposes.
The memorandum says spending on retention promotions rose from €134.48m in 2020 to €282.67m in 2025. It says those figures do not establish that promotions cause more gambling or losses and cannot alone identify people at risk.
Operators would need to adapt their information systems, calculate the model's variables, run it monthly and integrate the results into safer-gambling procedures. The resolution would take effect four months after publication in Spain's official gazette, the BOE. The regulator would periodically assess the mechanism's effectiveness and could update its variables and thresholds.