Introducing RAD-S1: a System One model for detecting AI agents that look human
Resemble AI launches Resemble Agent Detection, powered by its RAD-S1 model, to spot AI agents on websites, including disguised ones. In internal tests over 2,000+ visits it caught at least 98.8% of disguised agents, though by default it flags about 5 to 6% of human visits. The first million sessions are free.
What the article says
In a post dated 29 October 2026, Resemble AI notes that it is getting harder for a website to tell a human visitor from an AI agent, especially since these agents use real browsers, sometimes inside a customer's logged-in session. The company launches Resemble Agent Detection, built on the RAD-S1 model, which analyzes how a visitor moves the mouse, clicks, types and scrolls, then classes non-human visitors as declared, undeclared or disguised.
Integration uses a script of about 25 KB, with scoring done on Resemble's servers; typed text is not collected. In an internal test of more than 2,000 agent visits, RAD-S1 detects at least 98.8% of disguised agents, versus 3.1% to 14.8% for six compared tools. The default setting wrongly flags roughly 5 to 6% of human visits. The first 1 million sessions are free, and native mobile apps are not yet supported.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Resemble AI. The content and rights belong to the publisher: please read the original article.