Attackers sent 414,210 visits to an online marketplace posing as a widely used AI agent, with every visit targeting the sign-in page and returning an error, Help Net Security reports. The case points to a practical security lesson: watch what automated traffic does, not just what it calls itself.
Watch Desk analysis
What happened
Help Net Security, citing security company Cequence, describes a credential-stuffing operation that used the name of a consumer AI agent. The attackers tried stolen username and password combinations against the marketplace. The report says the fake traffic outnumbered genuine agent visits Cequence counted across five industries that week.
The same report describes a genuine personal agent that accessed a telecom carrier’s websites while presenting itself as an ordinary desktop browser. Its traffic grew from roughly 100 requests a day to 19,000 to 32,000 within three weeks. After a browser update, that traffic rose from almost none to about 98% adoption in around 48 hours, a pattern the report attributes to an operator updating a fleet of cloud browsers.
In a separate sample of 281,627 agent visits across five industries, 85% involved reading pages or comparing products. About one in seven involved more consequential actions such as booking, buying, signing in or managing an account.
Why it matters
A label is a poor security boundary when an agent can look like a browser and an attacker can claim to be an agent. The activity described here ranges from harmless browsing to account access and purchases, so treating every automated visit the same way risks either missing abuse or making legitimate services unnecessarily difficult to use.
Cequence recommends monitoring session behaviour, blocking patterns such as repeated failed visits to sign-in pages, and adding friction at sensitive moments such as payments or changes to account control. The report also says conventional bot scores missed many unauthorised login attempts in a separate example, while behavioural analysis identified them. These are reported cases, not evidence that one detection method will catch every attack.
Our read
This is a useful warning for anyone building services that welcome AI agents: identity signals alone are not enough. A sensible defence asks what a session is doing and raises the bar when it reaches for something consequential. The aim is to catch the credential-stuffing crew without making every legitimate assistant prove it is not a robot in a very small hat.
What to watch
- Whether more services distinguish routine browsing from sign-ins, purchases and account changes.
- How security teams detect agents that behave like ordinary browsers.
- Whether behavioural checks can reduce abuse without blocking legitimate agent activity.
Discussion spark: Should services let AI agents browse freely but require extra checks for sign-ins and payments, or is that distinction too easy for attackers to exploit?
Sources and evidence
- Source update (12 November 2026, 05:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.