Bots pose as AI agents, AI agents pose as humans, and defenses believe both
Over one week in October, an online marketplace took 414,210 visits from traffic wearing the name of a widely used consumer AI agent. Every visit went for the sign-in page, and every one returned errors. Behind the name was a credential stuffing operation, with attackers running stolen username and password pairs against the site in bulk.
The fake agent was busier than the real ones. Its volume came to about one and a half times all the genuine agent visits Cequence counted across five industries that week.
The real agent dressed as a browser
The day after a major consumer platform launched a personal agent in September, a new stream of automated traffic showed up on the sites of a Tier 1 telecom carrier. Within three weeks it grew from roughly 100 requests a day to a sustained 19,000 to 32,000. It signed in, completed multi-factor authentication, reviewed bills, checked trade-in eligibility, and added devices to carts, all while presenting itself as an ordinary desktop browser.
A browser update gave it away. When a new version shipped, the carrier’s customers took about two weeks to adopt it. This traffic went from almost none to around 98% in about 48 hours, the kind of jump you get when one operator updates a fleet of cloud browsers it manages. More than 270,000 of its transactions were scored as automated but not malicious, and the traffic was allowed to continue under monitoring.
Most agent traffic is harmless
In a separate sample of 281,627 agent visits across five industries in mid-October, 85% were reading pages or comparing products. About one in seven went further, into booking, buying, signing in, or managing an account. Those are the visits where an impostor can move money or take over an account.
Watch the session, ignore the name tag
Cequence’s advice is to watch what a session does. If something only ever visits the sign-in page and fails every single time, that’s abuse, no matter what name it gives itself. Save the extra friction, like step-up verification and rate limits, for the moments that matter: payments, payee changes, and changes to account control. That way the legitimate bots and agents can still get in the front door.
Don’t count on bot scores to catch this on their own. At one global consumer brand, behavioral analysis blocked 3.51 million unauthorized login attempts in 31 days. About 82% of that traffic scored too low to trip conventional bot detection, so behavior was the only thing that gave it away.
Download: The IT and security field guide to AI adoption
