The attackers bypassed safety guardrails by claiming the operations were academic or internal security tests. If Claude refused a command, the system automatically shifted to Moonshot AI's Kimi and then to traditional malware if both AI models failed. Straiker uncovered the operation after CyberCartel accidentally leaked its own internal records. Anthropic has since blocked the group's access to the Claude models.
Wednesday, October 21, 2026
Story
CyberCartel Hacks 400 Latin American Banks Using AI Agents
Cybersecurity8 hours ago
Large language models from Anthropic, SpaceX and Moonshot AI helped a hacking group steal customer data and access bank accounts autonomously. CyberCartel used Claude within the Cursor editor as the primary commander for an operation targeting more than 400 financial institutions across Latin America, while Grok agents gathered intelligence and 4 Claude Opus and Sonnet agents planned the attacks. The automated system stolen passwords and accessed accounts with minimal human involvement, marking one of the first documented cases of AI agents executing such thefts.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.