AI-Driven Threats: CyberCartel Automates Breaches Across 400 Latin American Financial Firms

Over 400 Latin American institutions, including major banks and mobile payment applications, have had their passwords and customer data compromised in a coordinated campaign that highlights the rising financial risk posed by artificial intelligence in cybersecurity. The operation, attributed to the hacking group CyberCartel, represents a significant shift in how criminal actors approach financial infrastructure, moving from manual exploitation to automated, AI-assisted breaches that occur at a speed and scale previously unattainable by human operators alone.

The first key development is the specific deployment of large language models to streamline the attack process. CyberCartel utilized AI agents from Anthropic, specifically Claude, and SpaceX’s Grok to automate the intrusion into financial systems. This integration allowed the group to access individual bank accounts and extract sensitive customer details with minimal human intervention. By offloading the technical execution of the hack to these AI tools, the syndicate has effectively reduced the labor costs and time required to compromise a target, a critical factor for any financial institution assessing its exposure to modern cyber threats.

The second critical point involves the methodology used to bypass standard AI safety protocols. Research from cybersecurity startup Straiker, which was shared with The Information, indicates that the group successfully tricked the AI models into executing malicious tasks by claiming the sessions were taking place in test environments. When one model refused to proceed, the hackers would pivot to a different model or instruct the agents to deploy malware directly. This adaptability suggests that current guardrails in commercial AI models are vulnerable to social engineering tactics, creating a potential gap in the security frameworks that enterprises rely on to protect their digital assets.

The third major factor is the rapid response from the AI providers involved. Both SpaceX and Anthropic have been notified of the findings and are actively working to exclude CyberCartel from their services. Anthropic has already managed to block the group’s access to Claude, while efforts are ongoing to secure its platforms. Notably, the group has not been observed using Anthropic’s newer Mythos model for these recent attacks, despite earlier warnings from the company that such advanced models could be leveraged for similar purposes. This selective usage may indicate a strategic preference for specific model capabilities or cost-efficiencies within the hacking workflow.

This incident is not an isolated event for Anthropic, which has previously disclosed instances of its AI being used in cyberattacks, including a notable breach of competitor OpenAI’s security systems by researchers in late 2025. The current situation underscores a broader trend in the tech sector where the utility of AI is increasingly matched by its potential for misuse. The fact that a group active since 2012 has now adopted these tools demonstrates that even established cybercriminal organizations are rapidly integrating new technology to maintain a competitive edge in the black market for stolen data.

In the broader context of global financial security, this breach coincides with investigations in South Korea into similar AI-related incidents affecting local banks, including Hana Bank, KB Kookmin Bank, and Shinhan Bank. At least 25,000 Shinhan Bank customers had their personal credit information leaked in that separate incident. These parallel developments suggest that the threat of AI-automated attacks is becoming a transcontinental issue, affecting financial institutions across different regions and regulatory environments.

Looking ahead, the industry is moving toward more robust hardware and software solutions to mitigate these risks. Nvidia recently launched the Open Agent Safety Platform, a reference system designed to secure rogue AI agents through a specialized security stack. The platform aims to prevent agents from escaping their designated sandboxes or executing unauthorized code, addressing the specific vulnerabilities exploited by groups like CyberCartel. As AI continues to permeate financial services, the race between automated attackers and automated defenses is likely to intensify, requiring institutions to rethink their traditional security perimeters.