Resemble AI tool finds 98.8% of AI agents that hide in its test

Resemble AI, a company that makes AI detection models, released Resemble Agent Detection on 29 October 2026. The model reads cursor movement, typing rhythm and page steps. It identifies each website session as a person, a bot or an AI agent. In its tests on 8 website stacks, it found 98.8% or more of the agents that were made to hide. 6 bot tools found 3.1% to 14.8%.

How it works

The model does not stop traffic. The customer selects the step for each result. The customer can let the agent in. It can send the agent to an API (a set of rules that lets software speak to software). It can keep the agent out of sensitive areas or stop destructive steps. Visitors see no puzzles, and the script does not collect typed text or passwords.

The numbers

The 98.8% result comes from tests by Resemble AI, not from an external group. The tests included agents on stealth browsers and agents that took control of a clean browser session. The first 1,000,000 sessions are free. After that, the price is $1 for each 1,000 sessions.

What is not known

Resemble AI gave no external test and no number for incorrect alarms on persons. It does not name the 6 tools. Its CEO, Zohaib Ahmed, says that detection is only the first step before a company can set rules for agents.

Insight: Banks stop agents while vendors race to sort them

Citigroup says it does not let third-party AI agents use customer accounts. In the same month, Forter released agent controls for merchants, and 4 firms showed an open identity check for agents. Resemble gives the first 1,000,000 sessions free, which reads as a move to control the point where agent traffic is sorted. Monitor which tool a firm uses first, since it can set the rules.