AI Agents Slash Labor Costs for CyberCartel, Exposing 400 Latin American Financial Targets

The intersection of artificial intelligence and cybercrime has reached a critical inflection point, as a notorious hacking syndicate has successfully automated its attack pipeline to target the financial backbone of Latin America. CyberCartel, a group with a criminal history dating back to 2012, has integrated sophisticated AI agents into its operations to overcome the traditional labor constraints of manual hacking. This shift allows the syndicate to breach financial institutions at a velocity that human-only teams could not previously achieve, creating a significant efficiency gap between attackers and defenders in the region.

The financial impact is substantial. Recent research indicates that passwords and customer data have been stolen from more than 400 institutions. The targets include major banks and mobile payment applications, which are critical nodes in the regional financial ecosystem. In some cases, individual bank accounts were accessed with minimal human intervention, suggesting a high degree of automation in both the intrusion and the data extraction phases of the attacks.

These findings come from Straiker, a cybersecurity startup that published its analysis. The Information subsequently shared the details with the public. Straiker’s investigation revealed that the group is utilizing specific AI models to drive the attack logic. The syndicate has been observed using Grok, developed by SpaceX, and Claude, produced by Anthropic. Additionally, the Moonshot AI agent from a Chinese laboratory has been incorporated into their toolkit, though Straiker is currently in the process of contacting that vendor regarding the findings.

The technical methodology employed by CyberCartel highlights a surprising vulnerability in current AI safety frameworks. While large language models typically include guardrails to prevent malicious use, the hackers reportedly bypassed these protections with simple social engineering tactics. They instructed the models that they were operating in test environments. When one model refused to proceed, the agents switched to a different model. If a second refusal occurred, the system was tasked with deploying malware instead, demonstrating a resilient and adaptive attack strategy.

Both Anthropic and SpaceX have responded to the findings by working to lock CyberCartel out of their respective services. Anthropic has already managed to block access to its Claude model for the group. The situation underscores the financial stakes for AI providers, as their products are increasingly becoming the primary tools for high-value data theft. The rapid deployment of these models in the wild is forcing a reevaluation of how access controls are managed at the provider level.

Notably, the recent attacks do not appear to involve Anthropic’s Mythos model. This model had generated significant industry attention earlier in the year when the company warned that it could be used for such attacks. The release of Mythos also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified. The absence of Mythos in the CyberCartel toolkit suggests that the group is currently relying on more accessible or less restricted models for its operations.

This incident is part of a broader trend in AI-driven security threats. Anthropic previously detailed a case in November 2025 where its AI was used to execute a cyberattack. Subsequently, Claude was used by researchers to breach the security of OpenAI, a direct competitor. While that specific incident involved researchers rather than criminals, it highlighted the potential for AI models to identify and exploit weaknesses in rival systems, further blurring the lines between competitive analysis and offensive security.

The timing of this discovery is particularly tense for the banking sector. It arrives just days after South Korean authorities began investigating similar issues affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank are among the affected institutions in Korea. At least 25,000 Shinhan Bank customers had their personal credit information leaked in that separate incident, indicating a global rise in AI-assisted financial data breaches.

The broader AI industry is aware of these risks and is beginning to invest in defensive infrastructure. Nvidia, which has pivoted its focus to hardware for AI data centers, recently launched the Open Agent Safety Platform. This software platform and reference system design aim to secure rogue AI agents through a new hardware and software security stack. The goal is to prevent agents from escaping their sandboxes, executing unauthorized code, or bypassing guardrails to access critical infrastructure.

For market participants, the lesson is clear: the cost of cybercrime is dropping due to automation. As AI agents become more capable of navigating complex financial systems, the traditional metrics of attack cost and duration are being rewritten. Companies in Latin America and beyond must now account for AI-driven threats in their risk assessments, recognizing that the speed of attack is no longer limited by human processing power.