Generative AI may bias crop protection advice away from agroecological approaches.

Generative AI holds ample potential to make agri-food systems more productive, resilient, and nature-friendly, but also poses risks. Here, following a methodical assessment of AI-generated advice on crop protection provided by ChatGPT-5, DeepSeek 3.1 up to 3.2-Exp, NormAI, KissanAI and VirtualAgronomist, we reveal systematic biases, faulty recommendations, and unethical machine behavior. As proprietary AI-powered advisories tend to favor commodified inputs and restrict the plurality of non-chemical alternatives, they may create informational biases that could, in turn, influence farmers' uptake of agroecological and biodiversity-driven solutions. For instance, whereas biological control, soil health and diversification tactics carry notable social-environmental benefits, they are systematically downgraded or even excluded by certain AI systems. Over time, such input-oriented foci can reinforce farmers' dependencies on purchased products - thereby eventually and inadvertently fueling pest proliferation, inflicting environmental harm, and eroding ecosystem services. Greater scrutiny, robust governance, and tailored regulations are urgently needed to safeguard environmental integrity and farmer livelihoods.