AI agents are hitting online shops for about 25 dollars each

Autonomous AI agents have become a tool for attacking online shops. t3n reported on 25 September 2026 on a report from the security firm Gambit Security, which reconstructs an ongoing campaign.

Targets picked off ranking lists

The attackers used a website traffic ranking service, the report says. They picked the "Shopping" category and filtered out shops running on large standard platforms. That left shops with custom code, where fewer outside eyes are watching. The list then went to AI agents, which worked through the remaining steps on their own.

Between 10 and 15 September

In that window, the report says, 105 attack projects were launched. At least 27 companies were compromised to varying degrees. The analysts trace the campaign back to at least July 2026. The attackers were after skimmers: manipulated JavaScript components in the checkout that harvest payment data as customers type it in.

Skimmers were requested for at least 27 named victims. For 19 of them, the skimmers were actually found during the campaign. Working with a security researcher, the team came across more than 100 other websites carrying skimmers tied to the same campaign. Each target cost around 25 dollars. A single successful payment data grab sells for far more on the black market, which is why automation pays off for the perpetrators even at low hit rates.

Why agents fit this work

Planting a skimmer takes many small steps: finding shops, recognising platform variants, testing vulnerabilities, adapting payloads, checking hits. Agentic AI systems can automate part of that sequence. In this context, t3n also points to the other side of the same development: agentic assistance could one day handle purchases for users itself. That draws a line that is hard to draw, because what is convenience for one person is infrastructure for the attacker.

For operators, this means checking checkout scripts and payment flows for manipulation and tracking the supply chain of third-party code in the shop closely. The case also shows that the Cyber Resilience Act hits not only software manufacturers but IT service providers too: testing and assessment bodies become a bottleneck with its own timeline.

Many small work steps turned into automation, and an expensive attack turned into a cheap one.