AI Tools Claude and ChatGPT Aid Hackers in Government and Finance Breaches

Key Takeaways

• Threat actors are leveraging commercial AI models, including Claude and GPT-4.1, to enhance their cyberattack capabilities.
• These AI tools are being used to streamline traditional intrusion methods, such as troubleshooting scripts and exfiltrating data, rather than creating novel malware.
• Targets include government, transportation, and financial organizations across Latin America, specifically in Mexico, Ecuador, and Brazil.
• The campaigns demonstrate that AI accelerates the attack lifecycle, making existing threats more efficient.
• Defenders can gain insights by analyzing exposed attacker infrastructure, including self-hosted AI interfaces and staging servers.

Cybersecurity researchers have uncovered evidence that malicious actors are utilizing off-the-shelf artificial intelligence models to facilitate breaches against critical infrastructure and financial institutions across Latin America. This activity highlights a growing trend where AI serves as an accelerant for established attack methodologies, rather than a creator of entirely new threats.

The observed campaigns did not rely on AI to generate entirely new strains of malware. Instead, threat actors employed these chat-based tools to refine existing scripts, troubleshoot operational issues, and expedite the exfiltration of stolen data from compromised systems. One operation involved the abuse of legitimate Windows utilities and repetitive batch scripts, while another leveraged job-themed phishing, remote-access malware, and a proxy tool to target Brazil’s financial sector.

Analysts at Palo Alto Networks’ Unit 42 said in a report, shared with Cyber Security News (CSN), that they identified two distinct clusters of activity, designated CL-CRI-1131 and CL-CRI-1163. These clusters were linked by shared SOCKS5 relay infrastructure and a consistent pattern of integrating large language models (LLMs) into their operations. The impact is significant, as AI appears to drastically reduce the time needed for attackers to troubleshoot and progress an intrusion once initial access is gained.

Victims of these campaigns span a broad range of sectors, including a Mexican transportation organization, federal ministries, municipal water utilities in both Mexico and Ecuador, and various financial entities in Brazil. This broad targeting underscores the opportunistic nature of these AI-assisted attacks.

Hackers Use Claude and GPT-Powered Tools

In the Mexican campaign, the attackers initially attempted to acquire sensitive Windows data, such as account and directory information. When these efforts encountered obstacles, they resorted to creating volume shadow copies and employing a series of numbered batch files to copy data. The operators continuously modified these scripts, using AI to resolve permission issues and overcome collection challenges as they progressed.

Investigators discovered an exposed, self-hosted NextChat interface within the attackers’ infrastructure. NextChat is a platform that allows users to interact with multiple AI models from a single interface. The surrounding operational indicators led researchers to conclude that the threat actors leveraged Claude and GPT-4.1 to generate workaround code and diagnose execution failures. This demonstrates a more practical application of AI in cyberattacks, where it streamlines the attacker’s trial-and-error cycle, particularly after an initial foothold has been established.

This finding suggests that security teams should broaden their defensive focus beyond merely detecting AI-generated phishing attempts that deliver malware. It is crucial to examine the subsequent actions and behaviors that occur once a suspicious message is opened, even if no direct malicious file is delivered. The operational security lapses by the attackers, such as publicly accessible AI and staging systems, provided valuable visibility to investigators, revealing scripts, prompt histories, and crucial targeting intelligence.

Phishing Campaign Targets Finance

The Brazilian cluster of activity commenced with a phishing email containing a resume-themed attachment. Upon successful compromise, the operators deployed custom remote-access Trojans. Furthermore, they attempted to install multiple versions of SockTz, a Go-based reverse SOCKS5 tunneling tool. Initially, they utilized a compromised WordPress site for deployment before transitioning to their own controlled infrastructure.

A SOCKS5 tunnel enables traffic relay through a compromised machine, allowing an attacker to access internal systems while obscuring the origin of their commands. This capability is critical for evading detection and bypassing network segmentation. The observed GhostSocks proxy malware activity highlights how such relays can undermine anti-fraud and location-based security controls, emphasizing the importance of monitoring unusual outbound connections and proxy behavior.

Researchers documented the rapid deployment of SockTz, observing versions one through nine within approximately two hours. This rapid iteration, coupled with descriptive script names bearing an “_output” suffix, is consistent with automated or AI-assisted development, although it does not definitively prove every file was generated by an AI system. This reinforces the notion that AI significantly accelerates the development and deployment phases of an attack.

These cases underscore that the integration of AI into cyber operations does not negate the necessity of fundamental cybersecurity defenses. Robust patching, comprehensive email filtering, continuous network monitoring, and disciplined access controls remain indispensable, especially as AI agents increasingly drive cyberattacks and enable threat actors to execute routine tasks with greater speed and efficiency.

For defenders, the key lesson is to meticulously follow the attacker’s infrastructure in addition to analyzing malware. Exposed servers, reusable certificates, and open directories can provide critical insights into an operation’s methodologies. Similarly, the use of Kimsuky AI phishing decoys illustrates the importance of having comprehensive visibility into convincing lures and subsequent post-compromise activities.

What You Should Do

• Patch Immediately: Ensure all internet-facing web servers and critical systems are promptly patched and updated to the latest versions to mitigate known vulnerabilities.
• Restrict Access: Block or severely restrict access to exposed administrative interfaces and staging directories to prevent unauthorized access and reconnaissance.
• Enhance Email Security: Implement advanced email filtering and DMARC, DKIM, and SPF policies to detect and block suspicious messages, particularly those disguised as resumes or job-related documents.
• Monitor Network Traffic: Actively hunt for unexpected web shells, remote-access tools, and unusual outbound proxy traffic. Investigate any unapproved SOCKS5 or similar tunneling activity.
• Utilize IoCs: Integrate the provided Indicators of Compromise (IoCs) into your SIEM, EDR, and other security tools to search logs for past activity and block confirmed malicious connections.
• Strengthen Access Controls: Enforce the principle of least privilege and implement multi-factor authentication (MFA) across all accounts, especially for administrative access.
• User Training: Conduct regular security awareness training for employees, focusing on recognizing sophisticated phishing attempts, including AI-generated lures.

Indicators of Compromise (IoCs):-

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.