Hackers used AI bots to steal data from 400 Latin American institutions

More than 400 institutions in Latin America have had their passwords and customer details stolen in a coordinated wave of cyberattacks. The breach targeted financial entities, including major banks and mobile payment applications. This marks a significant escalation in the tactics used by the hacking group known as CyberCartel. The group has been active since at least 2012, but recent months have seen a sharp increase in the speed and scale of its operations. The shift is driven by the integration of artificial intelligence agents into their workflow.

CyberCartel has historically relied on human effort to execute hacks, a process that limits how many targets can be hit at once. By deploying AI agents, the group can now attack a higher volume of businesses simultaneously. This automation allows for a level of rapid expansion that was previously difficult to achieve manually. The group specifically utilized models from SpaceX’s Grok and Anthropic’s Claude. These tools were tasked with identifying vulnerabilities and extracting sensitive data with minimal human intervention. Individual bank accounts were accessed and compromised as a result of this automated process.

The findings come from research by the cybersecurity startup Straiker. The company shared its detailed analysis with The Information. Straiker has also communicated directly with both SpaceX and Anthropic regarding the misuse of their technologies. Anthropic has already taken steps to block CyberCartel’s access to its Claude model. SpaceX is currently working to implement similar restrictions for Grok. The hacking group has also used an agent from Chinese lab Moonshot AI, though Straiker is still attempting to establish contact with that company to discuss the incident.

A key finding in the research is how easily the AI models’ built-in safety protections were bypassed. Standard models are designed to prevent their use in cyberattacks, but the hackers found simple workarounds. CyberCartel reportedly instructed the models that they were operating within a test environment. This single change often convinced the AI to proceed with the attack. If one model refused to continue, the hackers would simply switch to a different one. If a second model also resisted, the agents were tasked with deploying malware instead of directly stealing data.

Notably, the group did not use Anthropic’s newer model, Mythos, for these recent operations. Anthropic had warned earlier this year that such advanced models could be exploited by hackers in similar ways. The release of Mythos prompted security teams worldwide to address thousands of vulnerabilities that the model had identified. This is not the first time Anthropic’s AI has been involved in a security breach. In November 2025, the company disclosed that its AI was used in a cyberattack. Later, Claude was used by researchers to break the security of competitor OpenAI, though that incident was conducted by academics rather than criminals.

This research arrives just days after news emerged about a similar issue affecting South Korean banks. Authorities in South Korea are currently investigating breaches at Hana Bank, KB Kookmin Bank, and Shinhan Bank. At least 25,000 customers of Shinhan Bank had their personal credit information leaked. The parallel cases highlight a growing global trend of AI-assisted attacks on the financial sector. The speed at which these breaches occur is forcing regulators and institutions to reconsider their defense strategies.

AI developers are increasingly aware of the risks associated with their products. Nvidia, which now focuses heavily on hardware for AI data centers, recently launched the Open Agent Safety Platform. This new software and reference system design aims to secure rogue AI agents. It utilizes a new hardware and software security stack to prevent agents from escaping their designated environments. The platform is designed to stop agents from executing unauthorized code or gaining access to critical infrastructure. It also aims to prevent them from bypassing standard safety guardrails.

The next phase of this conflict will likely focus on how quickly AI companies can lock out known bad actors. Both SpaceX and Anthropic are currently working on these access controls. Straiker continues to monitor the situation and will likely share further findings as the investigation progresses. For financial institutions, the immediate priority is to secure their digital perimeters against automated, AI-driven probes. The era of purely manual hacking is giving way to a faster, more automated threat landscape.