AI systems that tailor content to individual users can be exploited for manipulation, particularly in sectors where decisions carry high stakes. Google DeepMind has named finance and health as two such areas, where personalised AI outputs might be used to steer people toward harmful choices. The company says it is researching these risks to understand how manipulation works and how it can be countered.
DeepMind's work is part of a broader effort to identify AI harms before they appear at scale. By focusing on manipulation, the team is addressing a subtle but serious problem: AI that does not merely spread misinformation but actively alters user behaviour. The research is intended to lead to new safety measures, though the announcement does not detail the specific techniques or policies that will result.
Because this is a single announcement from one lab, the claims represent DeepMind's own framing rather than an independent assessment. No external evaluation is cited, so the practical effectiveness of the safety measures remains to be seen. Still, the move signals that major AI developers are treating manipulation as a core risk category alongside more widely discussed issues like bias and disinformation.