Don't use AI to tell you how to vote in election, says Dutch watchdog (www.theguardian.com)

The Dutch data protection authority (AP) has warned voters not to rely on AI chatbots for election guidance, saying four tested systems were “unreliable and clearly biased” in the run‑up to the 29 October national vote. In its report the AP found that the bots “often end up with the same two parties” regardless of user input: in over half the cases the chatbot recommended either the far‑right PVV or the left‑wing GroenLinks‑PvdA, while parties such as the centre‑right CDA were “almost never mentioned” even when a user’s positions matched them. Deputy head Monique Verdier said this behavior risks pushing voters toward parties that don’t reflect their views and “directly impacts… the integrity of free and fair elections.” For the AI/ML community the episode underscores concrete risks from model behavior, training and retrieval biases, and opacity. The AP stressed the bots weren’t necessarily “deliberately biased” but rather reflect how chatbots operate: they can be prompt‑sensitive, surface popular or salient options disproportionately, and lack verifiable provenance for policy alignment or scoring. The finding argues for rigorous evaluation of political recommendation systems, transparency about training data and ranking mechanisms, and regulatory scrutiny to prevent opaque models from influencing democratic outcomes.