California's generative AI platform Poppy: a model for European administrations?

From July 1, 2026, California generalizes the use of its generative AI platform Poppy across its entire administration, following a pilot phase opened on September 29, 2025, involving over 2,800 public agents from 67 state services. Poppy is described by the California Department of Technology (CDT) as a vendor-agnostic platform built by state public agents, providing a single interface to multiple language models including Claude (Anthropic), Gemini (Google), GPT (OpenAI), and Nova (Amazon). The platform is intended for office tasks like drafting documents, summarizing large volumes of data, and searching public policy rules, explicitly excluding automated administrative decision-making. Key safeguards include that data stays within California's trusted environment, integrated detection of personally identifiable information (PII), and a guarantee that models are not trained on public data. Separately, Governor Gavin Newsom announced a partnership with Anthropic offering a 50% discount on Claude for state and local agencies, which is distinct from Poppy's multi-vendor approach. The California case provides a documented precedent for European administrations, though not directly transposable due to GDPR and AI Act constraints, but offering lessons on vendor neutrality, data control, and a phased pilot method.