AI · November 15, 2026 · Moderate impact
Axiomic Labs releases X3Guard-150M prompt guard that runs on phones and in browsers
Axiomic Labs has released X3Guard-150M, a 150-million-parameter prompt guard trained on its base model GPT-X3. For each prompt it flags in one pass whether the prompt is harmful and whether it is adversarial. The company says it outperforms models many times its size and can run entirely on edge devices and in the browser, with response times in the tens of milliseconds on iPhone. The model card and an in-browser demo are available.
A classifier this small that runs on-device makes prompt filtering cheap enough to put inside client-side applications.