Protecting your company from AI attack tools like Artex
The EU Commission was hit by a cyber campaign that investigators link to the AI tool Artex, and several internal systems went offline. Companies of every size can use the case to check their own defences and their own use of AI.

What happened at the EU Commission
According to a report by Politico, picked up by heise online, a large cyberattack kept the EU Commission busy for weeks. A spokesperson confirmed an AI-generated cyber campaign, discovered on 22 October and confirmed on 26 October. The Commission says it has contained most of the attack, but the full extent of the damage remains unclear.
Staff felt the effects directly. The HR platform Sysper, which handles payslips, leave requests and working time, and the medical expense system MyPMO were taken offline as a precaution. Internal translation services failed as well. Salaries were paid, but many employees could not record hours or request leave.
An anonymous official initially feared that extensive personal data had been compromised. The spokesperson said that, as things stand, no personal information has leaked. Investigations are running and systems are returning step by step. The Commission was reportedly very reserved in informing its own staff.
What Artex is and why it makes attacks easier
Artex began as an open-source project, an autonomous penetration testing platform built on large language models and written in Go. An FBI warning, as reported by heise, says that attackers have misused it since at least August to automate several phases of an attack. These range from reconnaissance through exploiting misconfigurations to data exfiltration.
The tool turns goals written in natural language into commands for specialised AI agents. For this it calls external models such as Claude, OpenAI's GPT or Chinese models like DeepSeek. The FBI reports that attackers use the Chinese models most often, because they are cheap and show no recognisable safeguards. In individual cases Artex needed under three hours from first targeting to privileged access.
This lowers the skill needed for complex operations. The original developer removed the project from GitHub in October after reports of abuse, yet copies and a commercial variant still circulate. Researchers counted over 1600 publicly reachable instances in early November.
How do I protect my company against Artex?
The sources describe the tool and its speed, and they name no specific defences. The checks below follow from what the tool automates: reconnaissance, misconfigurations and privilege escalation. Whether they suffice in your case depends on your environment, and your IT security lead or an external adviser should judge that.
- Review internet-facing systems and remove services that nobody needs, since automated reconnaissance finds them quickly.
- Check configurations of cloud services, file shares and remote access for open or overly broad settings.
- Reduce privileged accounts, separate administrator rights and require multi-factor authentication where possible.
- Make sure logging and alerting can detect unusual activity within hours, not days.
- Test backups and restoration, so that a disabled HR or finance system does not stop daily work.
- Rehearse an incident plan that includes how and when staff are told what is happening.
The Commission case shows the cost of vague communication. Employees were told only of a general technical problem, and that fuelled criticism. A short, honest internal notice helps people work around outages and report anything odd.
Using warnings from CERT-EU and authorities
CERT-EU, the cybersecurity team serving EU institutions, identified the use of Artex in early November and warned publicly about the software. The FBI has issued its own warning with technical detail. Such notices often contain indicators that your IT team can compare with its logs.
Assign one person to follow warnings from CERT-EU, the FBI and your national security authority, and decide who reads them and who acts. Small companies without a security team can ask their IT service provider to report on relevant warnings regularly. A warning that nobody reads protects nobody.
Setting up your own AI use securely
Attackers use external models because they are cheap and unguarded. Your own staff may use AI tools too, often without approval. A clear policy on which tools are allowed, with roles and a usage overview, shows where company data goes. Questions of data protection are covered in our guide to GDPR-compliant AI, and the topic of local AI explains how models can run in your own network.
Agents that act on systems need limits. The same automation that helps attackers can help staff, but anything that writes to a system should wait for a human decision, as described under agents with approval.
souveraen.ai is a platform for company knowledge from clouditiv GmbH in Auerbach, Germany. It runs On Demand, hosted in Germany, or on-premise, and its catalogue includes local models, models hosted in Germany and EU models with zero data retention. Agents prepare tasks and wait for approval before writing anything. More on this is on the security page.


