October 16, 2026 ยท Press release
The TOPS Trap: Local AI Models Hallucinated Facts, Cited Real Sources, and Sounded Confident
AIToolworx today published The TOPS Trap, a reproducible study of whether language models running on local hardware stay inside the evidence they are given. The study tested six local model configurations on a Raspberry Pi 5. It found that citations were not proof, that one model hallucinated facts identically in every run, that an instruction the user never typed was added to the prompt in all 32 runs of two Qwen 2.5 configurations, and that who built a local AI model is not on the box. Every prompt, run, log and hash is published so the work can be checked.
The findings apply to the exact configurations tested, on one platform and two topic packets, not to all local AI. The study is not peer reviewed, and AIToolworx has a commercial interest in AI security consulting. Before publication, the author offered Raspberry Pi, Hailo, Alibaba/Qwen, DeepSeek, Meta and Mistral AI the chance to comment; none responded.
Media contact: James Childers, Founder, AIToolworx, james.childers@aitoolworx.com