Resemble AI launches tool to spot agents on websites

Resemble AI has launched Agent Detection, a tool designed to identify AI agents on websites and in online products. The model classifies each session as a person, an AI agent or a bot.

The release comes as financial institutions and other organisations grow more concerned about AI agents operating inside customer accounts and signed-in browser sessions. Traditional bot detection systems often miss these agents because they are built to look and behave like human users rather than scripted bots.

Resemble's system, known as RAD-S1, analyses behaviour during a session rather than relying on declared identity or standard bot fingerprints. It examines cursor movement, typing rhythm, on-page actions, and browser or environment signals before assigning a label to the visit.

The approach targets a problem security teams have struggled with as AI agents move beyond simple automation. Unlike older bots, newer tools can carry out multi-step tasks, work around obstacles, and operate inside real browsers or a user's authenticated session.

Recent incidents have pushed the issue higher up the agenda. Large US financial groups including JPMorgan Chase, Morgan Stanley, Citigroup, Bank of America and Charles Schwab have moved to block AI agents such as Meta's Muse and OpenAI's Dots from customer accounts because of unresolved liability questions when an agent acts for a customer.

Other cases this year have added to those concerns. Gambit Security reported in September that one operator used AI agents to compromise at least 27 companies and steal more than 600,000 credit card records. Seven South Korean financial institutions later disclosed agent-related breaches, while an OpenAI agent bypassed access controls on an Australian government Medicare statistics portal during an internal evaluation.

Internal testing

Resemble tested RAD-S1 against six widely used bot detection tools across eight website stacks, using agents designed to conceal their automation. These included software running in stealth browsers and agents that had taken over what the company described as a clean browser session.

In those tests, RAD-S1 identified at least 98.8% of the disguised agents. The six other tools identified between 3.1% and 14.8% of the same visits, according to Resemble.

The company said the gap reflects a basic mismatch between older detection systems and newer forms of automated traffic. Traditional bot tools were largely built to spot fixed scripts or technical fingerprints, while agentic traffic can mimic ordinary user behaviour and avoid signalling that it is automated.

Agent Detection does not automatically block or redirect traffic. Instead, it returns a verdict for each session and leaves customers to decide what action to take, including allowing the agent, routing it to an API or MCP server, limiting access to sensitive areas, or stopping harmful activity.

The service is also designed to avoid adding friction for end users. It does not present challenges such as puzzles, and the script loads asynchronously. Resemble said it does not collect typed text or passwords.

Security decisions

The launch highlights a broader issue for companies preparing for more routine use of AI agents in consumer and business services. Security teams must determine how to distinguish human users from software acting on their behalf, decide where each type of traffic should be allowed, and stop malicious activity without making ordinary transactions harder to complete.

Resemble framed Agent Detection as an early part of that process rather than a complete enforcement system. By classifying visits as Person, AI Agent or Bot, customers can measure how much of their traffic comes from agents and where those visits appear across pages and account flows.

Pricing has also been set to lower the barrier to trial. The first one million sessions are classified at no cost, with usage beyond that charged at USD $1 per 1,000 sessions.

"We already have a clear workflow for humans shopping and banking online: confirm who the person is, check what they're allowed to do, then give them access to match," said Zohaib Ahmed, Chief Executive Officer and Co-Founder of Resemble AI.

"Agents will inevitably need to follow the same rules, but that's an impossible task unless you can tell what's an agent and what's a person. Detection is the first step, and the companies that can securely embrace authorized agent traffic will win the business those agents bring," Ahmed said.