Google confirms: Unlocking your phone's bootloader breaks local Gemini features (www.androidauthority.com)

Google has clarified in ML Kit’s GenAI Summarization API documentation that Gemini Nano — the compact, on-device version of its Gemini models — will not run on Android devices with unlocked bootloaders. Developers and enthusiasts who unlock their bootloader (or use rooted/dev builds) can see a FEATURE_NOT_FOUND error when attempting to use local Gemini features. The confirmation makes explicit what many had observed: the API deliberately refuses to load local model execution on devices in an unlocked state. This matters because Gemini Nano’s appeal is local, low-latency inference and offline capabilities; blocking it on unlocked devices limits experimentation and testing on developer phones and could force reliance on cloud-hosted models. Google’s rationale follows long-standing OEM practice of disabling sensitive features when a device’s trusted state is altered—an anti-debugging/attack-surface mitigation that, while not surprising, has practical implications for reproducibility, app QA, and hobbyist development. Affected users should use a secondary locked device, relock the bootloader if possible, or default to cloud APIs when local inference is required.