November 20, 2026

Exposed Server Reveals AI Agent Running Ransomware Attacks for Just $4 Per Target

A man in a black hoodie engaged in cybersecurity work using multiple monitors indoors.

An exposed server linked to an affiliate of the Gentlemen ransomware group has revealed how artificial intelligence is being used to automate almost every stage of a cyberattack.

Cybernews researchers discovered 3.1TB of information allegedly stolen from more than 30 companies, alongside tools, scripts and records showing how an AI agent was being used to compromise systems, exfiltrate data and prepare ransom demands.

The investigation offers a rare look inside an AI-powered ransomware operation and demonstrates how generative technology could dramatically reduce the time, expertise and cost required to attack businesses.

According to the researchers, the AI token cost of targeting an individual company could be as little as $0.40 to $4.00. That figure does not include the cost of servers, stolen credentials and other supporting infrastructure, but it illustrates how cheaply large parts of an attack can now be automated.

“Ransomware has effectively turned into a passive revenue stream,” said Aras Nazarovas, a security researcher at Cybernews, who discovered the leaking server.

AI Agent Used Throughout Attack Process

The exposed infrastructure reportedly contained malicious payloads, exploitation tools, stolen company information and an AI-agent system capable of supporting the complete attack lifecycle.

Rather than relying on a human attacker to manually adjust scripts for each organisation, the AI agent could analyse individual environments and modify its approach accordingly.

Researchers said the system was used to conduct reconnaissance, identify exploitable weaknesses, establish remote access and remove information from compromised networks.

It could also analyse stolen data, assess its potential value and help determine how much money should be demanded from each victim.

“Most attacks start as AI prompts. The attacker simply provides the AI agent with a GitLab URL, username, and password, likely obtained from stealer logs or purchased from initial access brokers. The AI agent then adjusts and modifies the exploit scripts based on individual victim environments. After that, the threat actor only queries the AI with questions, directing it to perform criminal actions. The AI agent has direct access to attacker-defined skills for recon, exfiltration, reverse shells, as well as files,” Nazarovas explains.

By allowing multiple targets to be processed simultaneously, this model could enable a single criminal operator to manage a volume of attacks that would previously have required a much larger team.

Stolen GitLab Credentials Provided Initial Access

The operation examined by Cybernews focused heavily on GitLab, a platform used by organisations to store source code and manage software-development projects.

Attackers appear to have begun with valid usernames and passwords, potentially obtained through information-stealing malware or purchased from initial-access brokers.

Access to a company’s GitLab environment can be particularly valuable to cybercriminals. Private repositories may contain source code, system documentation and details about internal infrastructure.

They can also expose API tokens, SSH keys, passwords and other secrets when developers have inadvertently stored credentials within code or configuration files.

Once inside, the AI agent could reportedly examine the victim’s environment, select or adjust suitable exploits and generate customised scripts.

Cybernews found 86 AI-generated Python scripts configured with real victim endpoints. These were allegedly used for tasks including maintaining access, extracting databases, archiving stolen files, transferring data and attempting to conceal evidence of the intrusion.

More Than 30 Organisations Compromised

The 3.1TB of exposed information was linked to more than 30 businesses operating across a broad range of industries.

Victims reportedly included organisations in marketing, healthcare, consulting, compliance, property, software development, telecommunications, manufacturing and transportation.

The diversity of the affected companies suggests the attacker was pursuing opportunistic targets rather than focusing on one sector.

Any organisation with exposed credentials, vulnerable software or an insecure GitLab deployment could potentially have been selected.

The AI system was reportedly capable of running attacks against several companies at the same time, further reducing the amount of direct human involvement required.

Data Theft Replaces Traditional Encryption

The operation also reflects a wider change in ransomware tactics.

Traditional ransomware attacks commonly involve encrypting an organisation’s files and demanding payment for a decryption key. In this case, the affiliate reportedly concentrated on stealing sensitive information and threatening to publish, auction or sell it.

The AI agent was used to create “pressure dossiers” containing information intended to support extortion negotiations. It also assessed the likely value of the stolen data and calculated potential ransom demands.

Other possible ways of monetising the information included offering source code or confidential documents to competitors, auctioning files on criminal marketplaces and using stolen data to support phishing or SMS fraud.

This approach means businesses could face a serious extortion threat even if their backups and disaster-recovery systems prevent operational disruption.

Once confidential information has been extracted, restoring systems from backups cannot undo the breach.

AI Reduces Cost of Cybercrime

The reported token cost of between $0.40 and $4 per company does not represent the complete cost of running a ransomware campaign. Criminals still require hosting, infrastructure, access credentials and ways to receive or launder payments.

However, automating reconnaissance, scripting, data analysis and ransom preparation could significantly lower the cost of scaling an operation.

AI could also reduce the technical ability required of the person directing an attack. Instead of writing and modifying every script manually, an operator may be able to describe an objective and allow an AI agent to complete much of the work.

This development risks increasing the number of attacks faced by companies, as criminals can target more organisations without a corresponding increase in staffing or expenditure.

Companies Urged to Strengthen GitLab Security

The findings underline the importance of protecting development platforms and responding quickly when credentials are exposed.

Organisations should avoid storing passwords, access tokens and other sensitive secrets within source-code repositories, including repositories that are designated as private.

Credentials suspected of appearing in information-stealer logs or criminal marketplaces should be revoked and replaced promptly.

Companies should also monitor GitLab accounts for unusual behaviour, including bulk cloning of repositories, unexpected access from new locations, the deployment of unauthorised tools and suspicious activity involving continuous integration and deployment runners.

Access to sensitive development infrastructure can be restricted to trusted networks or approved virtual private networks where appropriate.

GitLab installations and other internally hosted software should also be updated promptly to address known vulnerabilities.

Network monitoring may help identify unusual outbound transfers, reverse-shell connections or attempts to move laterally between systems.

Human Oversight Remains Essential for Defenders

Although the attack relied heavily on automation, the exposed server also demonstrates that cybercriminal infrastructure can suffer from basic security failures.

The server’s misconfiguration allowed researchers to examine the operation and identify the techniques being used.

The findings were responsibly disclosed to Lithuania’s Computer Emergency Response Team and police before publication. The information was subsequently shared with international partners.

For security teams, the case illustrates how AI is changing the economics and speed of cybercrime. Defences built around identifying fixed, recognisable malicious scripts may become less effective when AI agents can generate or adapt code for each target.

Businesses may therefore need to place greater emphasis on behavioural monitoring, credential security, rapid patching, access controls and detecting unusual data movements.

The full Cybernews investigation is available online