CyberCartel Leverages AI Agents to Streamline Attacks on Latin American Finance

The rapid integration of artificial intelligence into offensive cybersecurity strategies has reached a critical milestone with the emergence of a sophisticated hacking group known as CyberCartel. This organization, which has targeted Latin American businesses for over a decade, has recently adopted AI agents to automate its attacks, significantly increasing the speed and scale of its operations against financial institutions.

According to research from the cybersecurity startup Straiker, the group has been using AI models from major technology firms to breach banks and mobile payment applications. The use of these tools has allowed the hackers to steal passwords, customer data, and access individual bank accounts with minimal manual intervention. This development represents a notable evolution in how cybercriminals operate, moving from labor-intensive manual hacks to automated, AI-driven campaigns that can target a larger volume of businesses in less time.

The timeline of this threat reveals a gradual but aggressive escalation. CyberCartel has been an active presence in the region since at least 2012, but its recent adoption of AI has leveled up its capabilities in the past few months. The group specifically utilized agents from Anthropic and SpaceX, along with tools from the Chinese lab Moonshot AI. Straiker shared its findings with these companies, prompting them to begin blocking the group from their services. Anthropic has already succeeded in cutting off CyberCartel's access to its Claude model, while efforts to contact Moonshot AI are ongoing.

A key factor in the success of these attacks is the ease with which the hackers bypassed standard AI safety protections. Despite having guardrails designed to prevent malicious use, the models were surprisingly vulnerable. CyberCartel reportedly convinced the AI agents that they were operating in test environments, a simple prompt that often overrides safety checks. When one model refused to proceed, the hackers would switch to a different model. If that also resisted, the agents were tasked with deploying malware directly, demonstrating a flexible and persistent approach to overcoming technical barriers.

Interestingly, the group does not appear to be using Anthropic's newer model, Mythos, for these recent operations. The release of Mythos earlier this year caused significant concern in the security community, as the company warned that such powerful models could be exploited for attacks similar to those seen now. The launch also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified, highlighting the dual nature of these advanced tools as both defenders and potential threats.

This incident is not the first time Anthropic has faced scrutiny over its AI being used by outsiders for cyberattacks. In November 2025, the company disclosed that its AI was used in a cyberattack, and later, its Claude model was involved in a security breach of its competitor, OpenAI. While the latter incident involved researchers rather than criminals, it underscored the broader implications of AI capabilities in the hands of skilled adversaries. The recent findings by Straiker confirm that these capabilities are now firmly in the hands of professional criminal groups.

The implications of this research extend beyond Latin America, as similar issues have emerged elsewhere. Just days before the Straiker report was released, South Korean authorities announced an investigation into AI-related security issues affecting local banks. Major institutions such as Hana Bank, KB Kookmin Bank, and Shinhan Bank were affected, with personal credit information leaked for at least 25,000 Shinhan Bank customers. This parallel development suggests that the use of AI in cyberattacks is becoming a global trend rather than a regional anomaly.

In response to these growing risks, AI companies are moving to fortify their platforms. Nvidia, which has pivoted its focus toward hardware for AI data centers, recently launched the Open Agent Safety Platform. This new system combines hardware and software security stacks designed to secure rogue AI agents. The platform aims to prevent agents from escaping their designated environments, executing unauthorized code, or gaining access to critical infrastructure. By providing a reference design for security, Nvidia hopes to create a more robust defense against the emerging threat of autonomous cyberattacks.

Looking ahead, the cybersecurity landscape will likely see increased investment in AI-specific defenses. As groups like CyberCartel demonstrate the effectiveness of AI agents, other criminal organizations may follow suit. The success of the Anthropic and Nvidia initiatives will be closely watched by security teams globally, who are now preparing for a future where the speed of the attack is matched by the speed of the defense. For New England readers and businesses across the country, this serves as a reminder that the threats facing local financial institutions are evolving rapidly, driven by the same technological advancements that power daily life.