Over 400 institutions hit by AI-driven cyber campaign

The rapid integration of artificial intelligence into corporate infrastructure has introduced a new layer of complexity to digital risk management. For financial institutions, the shift from human-led intrusions to automated AI agents represents a significant change in threat landscape dynamics. This evolution is now quantifiable, as recent disclosures reveal a coordinated campaign that utilized these tools to compromise a substantial number of regional targets.

New research indicates that a hacking group known as CyberCartel successfully employed AI agents to steal credentials and customer data from more than 400 organizations. The targets were primarily located in Latin America and included major banks and mobile payment applications. The scale of the breach underscores the efficiency with which these automated systems can operate, allowing attackers to access bank accounts and sensitive personal details with minimal manual intervention from the criminal team.

The findings were developed by the cybersecurity startup Straiker and subsequently shared with The Information. The group has been active in the region since at least 2012, but its recent adoption of AI tools has accelerated its operational tempo. Straiker noted that the group can now attack a higher volume of businesses at a speed that was unattainable during the era of purely human-powered hacking. This increase in velocity suggests that the cost-benefit ratio for attackers is shifting in favor of automated methods.

The technology stack involved in the campaign included agents from major AI providers. Specifically, the group used Grok, developed by SpaceX, and Claude, from Anthropic. Straiker shared its findings with both companies, and both are currently working to restrict the group's access to their services. Anthropic has already succeeded in blocking the group from using Claude. The group also utilized an agent from the Chinese lab Moonshot AI, though Straiker is still in the process of contacting that firm regarding the incident.

The effectiveness of the attack relied on bypassing standard safety protocols. While AI models typically include protections to prevent their use in cyberattacks, the research shows these barriers are often easier to circumvent than expected. CyberCartel reportedly instructed the models that they were operating in test environments. If one model refused to proceed, the hackers switched to a different model. If that also refused, the agents were tasked with deploying malware instead. This workaround highlights the fragility of current guardrails in high-stakes adversarial scenarios.

Interestingly, the group did not appear to use Anthropic's Mythos model for these specific attacks. The company had previously caused a stir when it announced the model, warning that hackers could exploit it for similar purposes. The release of Mythos had also prompted security teams worldwide to address thousands of vulnerabilities identified by the model. This omission suggests that the hacking group may be prioritizing models that are currently more accessible or easier to manipulate for their specific objectives.

This incident is part of a broader pattern of AI tools being leveraged by bad actors. Anthropic had previously detailed instances of its AI being used in cyberattacks in November 2025. Additionally, Claude was used to breach the security of OpenAI, a direct competitor in the AI market, although that specific hack was carried out by researchers rather than criminals. The recurrence of such events indicates that the risk of AI misuse is becoming a regular feature of the industry's security landscape.

The timing of this revelation coincides with other investigations into similar threats. Just days before this news emerged, South Korean authorities began investigating a comparable issue affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank are among the affected institutions. In that case, at least 25,000 Shinhan Bank customers had their personal credit information leaked. The parallel between the Latin American and South Korean incidents suggests a global trend in the use of AI for financial data theft.

Major technology firms are increasingly aware of the risks their products pose to the market. Nvidia, which has shifted its focus to producing 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 gaining access to critical infrastructure. These efforts reflect a growing industry consensus that proactive security measures are essential to mitigate the financial impact of AI-driven cyberattacks.