Attackers Leverage AI Tools in Latin American Campaigns for Data Exfiltration
Two distinct cyberattack campaigns targeting organizations in Latin America are utilizing artificial intelligence to enhance their data exfiltration capabilities, according to a new report.
Palo Alto Networks' Unit 42 has identified two separate, ongoing cyberattack campaigns that are actively employing artificial intelligence (AI) to target organizations in Latin America, primarily for the purpose of data exfiltration. The campaigns, tracked as CL-CRI-1131 and CL-CRI-1163, exhibit distinct geographic focuses and victimology but share overlapping infrastructure and a common reliance on AI to streamline operations.
The first campaign, CL-CRI-1131, has impacted a transportation organization, along with federal government ministries and municipal water utilities in Mexico and Ecuador. This operation predominantly utilizes living-off-the-land (LotL) techniques, with attackers orchestrating iterative batch scripts to manipulate and exfiltrate sensitive data. Notably, the threat actors self-hosted NextChat instances on compromised operational infrastructure, indicating a deliberate effort to leverage AI-powered chatbots for command and control or analysis.
The second campaign, CL-CRI-1163, has focused on the Brazilian financial sector. This activity represents an expansion of previously observed targeting of vulnerable web servers through a job-themed phishing campaign. Attackers in this cluster have deployed custom remote access Trojans (RATs) and tunneling tools, including a Go-based SOCKS5 proxy. The iterative naming convention of these tools suggests potential AI enablement in their development or operation.
Both campaigns demonstrate a significant trend in the evolving threat landscape of Latin America. The shared SOCKS5 relay infrastructure and the explicit use of AI, particularly commercial large language models (LLMs), to orchestrate operations highlight a sophisticated and coordinated approach by diverse threat groups. This indicates a broader evolution beyond isolated incidents, with regional actors independently adopting advanced proxy networks and AI integration to enhance their attack efficiency.
Analysis of CL-CRI-1131 revealed the attackers' trial-and-error process during an April 2026 compromise, consistent with the use of LLMs for script generation and troubleshooting. The attackers struggled with initial attempts to dump critical system files like the Security Account Manager (SAM) registry hive and the domain controller NTDS.dit file. Their iterative script fixes and permissions checks suggest a learning or refinement process likely guided by AI assistance.
Further investigation into the exfiltration infrastructure for CL-CRI-1131 uncovered an active Let's Encrypt TLS certificate associated with the domain m-doxa-apodo.duckdns[.]org. Searching for the 'm-doxa' prefix revealed that attackers established this infrastructure in February 2026, initially using a single certificate. As the operation evolved, they rotated their infrastructure and generated new multi-Subject Alternative Name (SAN) certificates in April and June 2026, indicating a dynamic and adaptive approach to maintaining command and control and facilitating data exfiltration.
The technical and behavioral overlaps between these two distinct clusters, despite their different targets and primary techniques, underscore a significant shift in threat actor methodologies. The integration of AI tools, whether for script generation, operational planning, or enhancing existing malware capabilities, represents a growing challenge for cybersecurity defenders. The ability of threat actors to independently adopt and integrate these advanced technologies signals a more formidable and rapidly evolving threat environment in the region.
Palo Alto Networks customers are protected against these threats through products like Advanced WildFire, Advanced URL Filtering, Advanced DNS Security, and Cortex XDR/XSIAM. The Unit 42 team also offers incident response services for organizations that suspect a compromise.