More than 9,000 requests per minute and renewed attacker activity less than an hour after the affected systems came back online. An incident at a European e-commerce platform shows how quickly an automated attack can escalate. Eye Security used AI agents to help analyse the incident, while its security team decided how to respond.
Berlin, 16 November 2026 – A major European e-commerce platform was the target of a suspected AI-assisted cyberattack, according to Eye Security. The attackers gained access to the platform’s IT infrastructure, stole credentials and used them to target other systems. Eye Security suspects that the attackers used ARTEX, an AI-powered attack tool. The company believes the motive was financial gain, with gift vouchers among the likely targets.
Initial measures fail to stop the attack
According to Eye Security, the attack began with a vulnerability in a custom-built upload feature. The attackers used a stolen API key within minutes of obtaining it and adapted their activity within minutes of an access restriction being put in place. At its peak, the platform recorded more than 9,000 requests per minute.
Less than an hour after the affected systems came back online, security alerts were triggered again. The attackers still had access to the source code repository and had tampered with the automated software deployment process. As a result, unauthorised applications were automatically redeployed after being removed. Bringing the systems back online had not cut off every route the attackers could use.
AI speeds up analysis, but people make the decisions
Eye Security used its own AI agents to analyse security data in parallel and identify suspicious activity. The agents had read-only access to the customer’s environment. The security team retained control over any changes to the systems.
A mistake during the investigation showed why the agents’ findings still needed to be checked. One agent misidentified an employee’s usual network connection as a guest Wi-Fi connection. Other agents repeated that assessment, and the error was only corrected when the finding was checked against up-to-date data. In this case, agreement between several agents did not amount to independent confirmation.
After the second alert, Eye Security and the platform operator isolated the affected environment completely. The team assessed the extent of the compromise and replaced affected keys and passwords. After several days of work and the introduction of additional safeguards, the platform was able to resume most of its operations. Eye Security says it detected no further successful attacks on the systems.
What organisations can learn from the incident
Fixing the original vulnerability does not mean an attack is over. Before bringing systems back online, organisations should check whether compromised credentials are still valid, whether attackers have established other routes into the network and whether the automated deployment process is back under their control. It is just as important to agree responsibilities in advance.
‘If you wait until an incident to decide who makes the calls and which systems can be taken offline, you lose valuable time,’ says Fabian Lochmann, Cybersecurity Evangelist at Eye Security.


