Attackers used an AI agent name for 414,210 marketplace visits

Cequence, a company that makes tools to keep apps safe, counted 414,210 visits to an online marketplace in one week of October. The visits used the name of a known consumer AI agent, but each one went to the sign-in page and gave an error. Cequence tells that attackers used stolen username and password pairs, at about 1.5 times the volume of all other agent visits that it counted.

The reverse case

A large consumer platform started a personal agent in September. A Tier 1 telecom carrier then saw new traffic become more, from about 100 requests a day to as much as 32,000. It was able to sign in, did multi-factor checks and added devices to carts, but it told the carrier that it was a desktop browser.

The numbers

• In a sample of agent visits, 85% only read pages or compared items.
• About one in seven went on to book, buy, sign in or manage an account.
• At one global brand, bot detection did not find about 82% of the login tries that behavior checks stopped.

What is not known

Cequence did not name the agent or the platform. It let the carrier traffic continue and monitored it, as automated but not dangerous. It gave no number for how many agent names are not correct.

Insight: Name tags and bot scores both fail

Cequence saw an agent name that was not correct do a login attack. It also saw an agent of a platform that told the carrier it was a desktop browser. Each agent or attacker can use a name that is not correct for it, and thus the steps of a session are the only sign left. Leaders get more from logs of each step than from bot scores.