Resemble AI Agent Detection Spots Agents Acting Human

Resemble AI has launched a new tool designed to answer a question that is becoming urgent for nearly every website owner: is the visitor on the other side of the screen a person, or an artificial intelligence agent pretending to be one? According to the company, its new Resemble AI Agent Detection product, built on a model called RAD-S1, classifies each session as a person, a bot, or an AI agent based on how the visitor behaves, rather than on what the visitor claims to be.

The launch, announced on October 29, 2026, arrives at a moment when agent traffic is no longer a curiosity. Personal agents can now browse, shop, and manage accounts on behalf of their owners, often inside a real browser or a signed-in session, and often without identifying themselves. For businesses, that creates a blind spot: they cannot govern traffic they cannot see, and they cannot safely welcome helpful agents while blocking harmful ones.

According to the announcement, reported by EIN Presswire, the system analyzes cursor movement, typing rhythm, and on-page actions alongside browser and environment signals. Because the verdict rests on behavior, the company says it can identify agents that run in real browsers and even disguise their identity, including agents built to conceal their automation.

Why Resemble AI Agent Detection Arrives Now

The timing is hard to miss. According to the announcement, the launch comes a week after major banks including JPMorgan Chase, Morgan Stanley, Citigroup, Bank of America, and Charles Schwab moved to block AI agents from customer accounts, citing unresolved questions about liability when an agent acts on a customer's behalf. Banks are not alone in their caution, and the broader debate over agent access to financial accounts has already reached other platforms, as seen in recent coverage of how agentic tools are entering everyday service work.

The announcement also points to a string of incidents this year that show what agents can already do. According to the company, Gambit Security reported in September that a single operator used AI agents to compromise at least 27 companies and steal more than 600,000 credit card records. Weeks later, seven South Korean financial institutions disclosed agent-related breaches. In Australia, an OpenAI agent bypassed access controls on a government Medicare statistics portal during an internal evaluation, an incident that reportedly went unreported for months.

Those cases illustrate a split that matters for the whole agent ecosystem. Helpful agents promise faster checkouts and simpler account management, while malicious agents can move at machine speed through systems designed for humans. Without a reliable way to tell the two apart, many companies have chosen the bluntest option: block agent traffic altogether. That protects the site, but it also shuts out legitimate agents and the customers who use them.

How The Behavioural Model Works

Traditional bot detection was built for a different era. Old bots announced themselves or followed fixed scripts that left recognizable fingerprints. AI agents, by contrast, are designed to complete multi-step tasks and work around obstacles, which makes them far harder to spot with signature-based tools. Resemble AI says its model was built specifically for this gap.

In internal testing across eight website stacks, RAD-S1 identified at least 98.8 percent of agents built to conceal their automation, according to the company. That included agents running on stealth browsers and agents that took over a clean browser session. Six widely used bot detection tools tested on the same visits identified between 3.1 percent and 14.8 percent, according to the announcement. Those figures come from the company's own testing, so independent verification will matter as the product reaches more sites.

Importantly, the product does not block or route traffic on its own. Customers decide how to act on each verdict, whether by allowing an agent, redirecting it to an application programming interface or MCP server, restricting it from sensitive areas, or stopping destructive actions. Teams can set responses page by page and pair verdicts with existing identity systems or emerging standards such as Sierra's Personal Agent Protocol.

Visitors encounter no puzzles or challenges, according to the company. The script loads asynchronously and does not collect typed text or passwords, which is meant to keep the experience smooth for genuine human visitors while still giving security teams visibility into agent traffic.

What This Means For Agents And The Sites They Visit

For the agent ecosystem, detection is a double-edged development, and that is exactly why it matters. On one hand, a reliable detection layer could make sites more willing to welcome agents, because they can measure agent traffic, route it to proper interfaces, and set clear rules instead of banning it outright. Agents that identify themselves and behave well could gain access that is currently closed to them.

On the other hand, detection raises the bar for agents that rely on blending in. As more sites deploy behaviour-based classification, the era of agents quietly passing as humans may be closing, and agent builders will need to think harder about transparent identification, delegated permissions, and auditable behaviour. That shift echoes the wider conversation about governed agent access, including the debate over shutdown controls for risky AI systems.

Resemble AI's chief executive, Zohaib Ahmed, framed detection as the first step toward a workable set of rules for agents online. According to the announcement, he said companies already have a clear workflow for humans shopping and banking online, confirming who the person is, checking what they are allowed to do, and then giving them matching access. Agents, he argued, will inevitably need to follow the same rules, but that is impossible unless sites can tell what is an agent and what is a person.

The product is available now, according to the company, with a free tier covering the first one million sessions after adding a single script tag, and pricing beyond that at one dollar per one thousand sessions. Whether rival detection vendors respond with their own behavioural models will be worth watching, because the first company to define agent traffic measurement may also shape how the rest of the industry talks about it.

For now, the launch marks a turning point in how the web treats its newest visitors. Agents are no longer rare enough to ignore, and they are no longer invisible enough to go unmeasured. The question for every site is no longer whether agents are visiting, but what to do once they are spotted. Sources: EIN Presswire announcement from Resemble AI and Resemble AI.