Overview
Severity: CRITICAL | Affected: Nexus AI | Category: incident
Nexus AI, a leading developer of generative models, disclosed a major security breach originating from a highly targeted social engineering attack. Attackers used a deepfake voice clone of the company's CEO in a vishing (voice phishing) call to a senior engineer, convincing them to bypass a multi-factor authentication prompt. Once inside the corporate network, the threat actors moved laterally and gained access to a development environment containing the pre-release weights for Nexus's upcoming 'Aura-5' language model. The company's investigation also revealed that a sanitized, but still sensitive, subset of user interaction data used for Reinforcement Learning from Human Feedback (RLHF) was exfiltrated. Nexus AI has since invalidated the compromised credentials, rotated all production keys, and is working with law enforcement. The incident highlights the growing threat of AI-powered social engineering and the critical importance of protecting model IP and training data.