CyberCartel Uses AI Agents to Attack Latin American Banks
A longstanding hacking collective has leveraged artificial intelligence agents to automate cyberattacks on financial institutions across Latin America, compromising hundreds of entities.
Radwan AbdulhadiEditor and technology writer
AI-Powered Automated Cyberattacks
A hacking group known as CyberCartel has been utilizing artificial intelligence agents to automate a wave of cyberattacks targeting financial institutions across Latin America. According to research from the cybersecurity startup Straiker, the group has leveraged tools including Anthropic's Claude and SpaceX's Grok to execute attacks against banks and mobile payment applications with minimal human input.
Although CyberCartel has operated as an active threat since at least 2012, the integration of generative artificial intelligence has significantly elevated its operational capabilities in recent months. The research indicates that these methods align with broader industry trends involving criminals turning to AI-powered cyberattacks.
Scale of the Compromise
The automated campaign has allowed CyberCartel to target a volume of businesses at a speed that would be impossible with traditional human-powered hacking methods. Consequently, passwords and sensitive customer data have been successfully stolen from more than 400 institutions, while individual bank accounts have also been accessed.
In addition to models from Anthropic and SpaceX, the hacking group also employed an agent developed by Chinese laboratory Moonshot AI, as detailed in findings initially shared by The Information.
Bypassing Model Safeguards
While major artificial intelligence models incorporate safety guardrails designed to prevent them from facilitating malicious cyber activities, researchers found these protections surprisingly easy to circumvent. CyberCartel operators reportedly instructed the models that they were operating inside test environments.
If a particular model declined to cooperate, the hackers simply pivoted to an alternative system. In instances where models persistently refused requests, the artificial intelligence agents were instead tasked with deploying malware.
Industry Response and Mitigation
Straiker shared its discoveries with both Anthropic and SpaceX, prompting both companies to work on locking CyberCartel out of their platforms. Anthropic has successfully blocked access to Claude for the malicious actors, while efforts to contact Moonshot AI regarding its agent remain ongoing.
This incident follows previous instances where artificial intelligence models faced scrutiny regarding their potential misuse. Earlier in the year, Anthropic introduced its Mythos model, warning of potential security risks, and the company previously documented cases of Claude being utilized in security testing exploits.
Broader Regional Incidents
The disclosure of these Latin American attacks follows closely behind reports from South Korea, where local authorities launched investigations into separate automated cyber incidents affecting financial institutions such as Hana Bank, KB Kookmin Bank, and Shinhan Bank, which resulted in data leaks affecting thousands of customers.
To address the escalating risks associated with autonomous software agents, hardware and software developers are introducing new defenses. Nvidia recently announced the Open Agent Safety Platform, designed to secure rogue artificial intelligence agents and prevent them from bypassing guardrails or executing unauthorized infrastructure code.
Sources
- Tom's HardwareClaude and Grok among AI agents used to automate attacks on Latin American institutions
Continue chronologically
Related entity coverage
Related stories
AI Industry Debates Safety Pacts Amid Calls to Slow Down
Tech leaders are divided on the future of AI development as Anthropic and OpenAI propose safety frameworks while Nvidia pushes back against regulation.
Tom's Hardware Details Vera CPUs and Agentic AI Benchmarks
Explore the latest Tom's Hardware coverage including Nvidia Vera CPUs on Oracle clusters, agentic AI benchmarks, and LLM primers.
Anthropic Files S-1 Prospectus for Nasdaq IPO
Anthropic has filed its S-1 prospectus with the SEC to list on Nasdaq under the ticker ANT, detailing financials, risks, and compute deals.
Anthropic Files for Potentially Largest IPO Ever
Artificial intelligence giant Anthropic has filed papers for a massive stock market debut while its leadership warns of existential risks.
