Bots pose as AI agents, AI agents pose as humans, and defenses believe both
Cequence Security found an online marketplace hit with 414,210 fake-agent credential-stuffing visits in one week in October, while a separate telecom carrier saw real AI-agent traffic from a major consumer platform's new agent jump from about 100 requests a day to over 19,000 within three weeks.
Cequence Security found that over one week in October, an online marketplace received 414,210 visits impersonating a widely used consumer AI agent — and every one of them went straight for the sign-in page and failed.
It wasn't the real agent at all — it was a credential-stuffing operation running stolen username-and-password pairs against the site in bulk, dressed up with the agent's name.
Separately, the day after a major consumer platform launched a personal AI agent in September, a new stream of automated traffic appeared on a Tier 1 telecom carrier's site. Within three weeks its volume grew from about 100 requests a day to a sustained 19,000–32,000.
That traffic signed in, passed 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-version update later gave it away: legitimate customers took about two weeks to adopt the new version, but this traffic jumped from almost none to about 98% adoption in roughly 48 hours — the signature of a single operator updating a managed fleet of cloud browsers.
Across a separate sample of 281,627 agent visits spanning five industries in mid-October, Cequence found 85% were simply reading pages or comparing products. About one in seven went further — into booking, buying, signing in or managing an account — which is where an impostor posing as an agent can actually move money or take over an account.
More than 270,000 of the telecom carrier's agent-like transactions were scored as automated but not malicious and allowed to continue under monitoring — exactly the traffic where a convincing impostor is hardest to tell apart from a real customer or a legitimate agent.
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 on conventional bot-detection systems to be caught by score alone, and was only flagged through behavior.
Cequence's recommendation: judge sessions by what they do rather than the name they give themselves. A session that only ever hits the sign-in page and fails every time is credential stuffing, whatever label it uses. Save extra friction — step-up verification, rate limits — for the actions that matter: payments, payee changes and changes to account control, so legitimate bots and agents can still get through the front door.
What happened
Cequence Security found that over one week in October, an online marketplace received 414,210 visits impersonating a widely used consumer AI agent — and every one of them went straight for the sign-in page and failed.
It wasn't the real agent at all — it was a credential-stuffing operation running stolen username-and-password pairs against the site in bulk, dressed up with the agent's name.
What changed
Separately, the day after a major consumer platform launched a personal AI agent in September, a new stream of automated traffic appeared on a Tier 1 telecom carrier's site. Within three weeks its volume grew from about 100 requests a day to a sustained 19,000–32,000.
That traffic signed in, passed 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-version update later gave it away: legitimate customers took about two weeks to adopt the new version, but this traffic jumped from almost none to about 98% adoption in roughly 48 hours — the signature of a single operator updating a managed fleet of cloud browsers.
Who is affected
Across a separate sample of 281,627 agent visits spanning five industries in mid-October, Cequence found 85% were simply reading pages or comparing products. About one in seven went further — into booking, buying, signing in or managing an account — which is where an impostor posing as an agent can actually move money or take over an account.
Why it matters
More than 270,000 of the telecom carrier's agent-like transactions were scored as automated but not malicious and allowed to continue under monitoring — exactly the traffic where a convincing impostor is hardest to tell apart from a real customer or a legitimate agent.
Technical details
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 on conventional bot-detection systems to be caught by score alone, and was only flagged through behavior.
What security teams should do
Cequence's recommendation: judge sessions by what they do rather than the name they give themselves. A session that only ever hits the sign-in page and fails every time is credential stuffing, whatever label it uses. Save extra friction — step-up verification, rate limits — for the actions that matter: payments, payee changes and changes to account control, so legitimate bots and agents can still get through the front door.