A University of Malta professor says AI could help flag grooming risks in online games by analysing player behaviour, while also warning that AI agents could be used to approach children. His proposal is an argument for protective tools, not evidence that grooming-detection systems are already proven.
Watch Desk analysis
What happened
In an interview with MaltaToday published on 29 October, Georgios N. Yannakakis said game platforms could train systems to flag behaviour associated with grooming, using data on how players interact. He suggested an AI companion that could warn players before they chat with a person or an AI agent.
Yannakakis cited a 2021 study, co-authored with Ubisoft researchers, in which models distinguished players sanctioned for toxicity from others with at least 82 per cent accuracy. That result concerned toxic behaviour, not grooming. He said reported grooming cases could provide training data, but the article gives no evidence that a model has been built or tested to detect grooming itself.
He also warned that generative AI could create convincing avatars that talk to children, and argued that studios developing AI features should invest in protective measures. MaltaToday’s account also notes that Roblox uses its Sentinel AI tool to detect early signs of child endangerment; the company says Sentinel identified nearly 70 per cent of cases detected in the 12 months to 7 August 2026.
Why it matters
The proposal puts a practical question on the table for game platforms: can behavioural signals help identify risky interactions early enough to protect children, without turning ordinary play into indiscriminate surveillance? The distinction matters. A model that identifies toxic players is not automatically a grooming detector, and the article does not establish how a proposed system would perform or guard against false alarms.
Our read
Yannakakis makes a timely case for treating safety as part of the AI development budget, not a feature to bolt on after the avatars have learnt to talk. The most useful next step is evidence: what signals a system uses, how well it detects actual grooming, and what happens when it gets a call wrong.
What to watch
- Whether game studios test behaviour-based grooming detection and publish evaluation results.
- What safeguards would limit monitoring, false alerts and the use of players’ data.
- Whether platforms explain how their existing safety tools perform and how cases are reviewed.
Discussion spark: Should game platforms use AI to flag behaviour that may signal grooming, even if that means monitoring more player interactions, or should child-safety tools be limited to reports and direct evidence?
Sources and evidence
- AI could be used to spot grooming from players’ in-game behaviour (29 October 2026, 00:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.