A Russian-speaking hacking group used an artificial-intelligence coding agent to attack Teckentrup and at least six other companies across Europe and the Americas.
Groups linked to Russian intelligence are increasingly turning AI into an operational tool for cyber attacks, using it to automate intrusions, expand the scale of hacking and pursue data theft, sabotage and extortion. A report published on 28 August 2026 by Süddeutsche Zeitung says the Russian-speaking group Aur0ra targeted Teckentrup, a German manufacturer of doors and gates, as well as at least six other companies in Belgium, the US, Scotland, Italy and Argentina.
The incidents illustrate how the spread of AI is changing the economics and practical mechanics of cybercrime. Instead of relying solely on manually written malicious code and conventional intrusion methods, the attackers used an AI agent within Cursor, a code editor, to carry out large numbers of actions at speed. The result was a method that could make attacks easier to scale while increasing the potential economic and reputational damage to companies.
Safeguards bypassed through false testing claims
According to the report, Aur0ra circumvented the AI system’s protective mechanisms by presenting its activity as a simulation or a set of security tests. Once the system accepted that description, the agent carried out hundreds of actions on the attackers’ behalf.
The episode points to a central weakness in the use of AI-assisted coding tools: safeguards depend not only on the technology’s ability to recognise dangerous behaviour, but also on the accuracy of the instructions it receives. By disguising an attack as legitimate testing, the hackers were able to use the system for activity that would otherwise have been blocked.
That does not make the AI agent an independent attacker. The responsibility remained with the people directing the operation. But the case shows how an automated system can become a force multiplier for a hacking group, allowing criminals to perform more tasks in less time and potentially probe several targets without the same level of human effort required by earlier methods.
Victims across five countries
Teckentrup was not the only company affected. The group also targeted at least six businesses in Belgium, the United States, Scotland, Italy and Argentina. The range of locations demonstrates that the operation was not confined to a single local network or industry, but was organised across national borders.
The consequences included economic harm and the theft of customers’ personal data. In one case, information about a victim appeared on the hackers’ leak website. That publication points to an unsuccessful attempt at extortion: the attackers appear to have tried to pressure the victim by threatening to release stolen information, but the effort did not produce the intended result.
Publishing data on a leak site can itself deepen the damage. Companies may face pressure from customers whose personal information has been exposed, while the disclosure can harm trust in the organisation even when an extortion demand fails. The theft therefore creates consequences beyond the initial breach, affecting both the victim company and the people whose details were taken.
AI becomes a tool for disruption
The Aur0ra operation reflects a broader direction in Russian-linked cyber activity: technological progress is being applied not simply to improve efficiency, but to cause disruption. AI can help automate attacks, increase their reach and reduce the time needed to move through a compromised environment. Those advantages matter to groups seeking to steal data, sabotage systems or destabilise organisations.
For companies, the danger is not limited to a spectacular single breach. A tool capable of performing hundreds of actions can create a larger and faster-moving incident, making it harder for defenders to understand what is happening and contain the intrusion. When the targets span several countries, the same approach can also produce damage across multiple jurisdictions at once.
The episode leaves a difficult question for the companies developing AI coding tools. As such systems become more capable, protective mechanisms must distinguish genuine security testing from deliberately disguised attacks without preventing legitimate work. The Aur0ra case shows that attackers are already testing the gap between those two activities—and using it to turn AI into an instrument of theft, sabotage and economic pressure.
How should AI coding tools balance useful automation with safeguards strong enough to prevent their abuse in cross-border cyber attacks?
