Overview
Severity: CRITICAL | Affected: Cognition Corp | Category: breach
Emerging AI leader Cognition Corp announced a critical security breach where attackers exfiltrated the proprietary model weights for its flagship 'Insight-5' LLM, along with terabytes of user prompt data. The breach, which occurred on February 12th, was traced back to a compromised Kubernetes service account with overly permissive access to the company's cloud storage buckets. This incident represents a new frontier in AI security threats, as the theft of model weights—the core intellectual property of the AI—allows attackers to replicate, analyze, or create malicious derivatives of the powerful model. Cognition Corp has notified affected users and is working with cybersecurity firms to investigate the full scope. The incident has intensified calls for stricter access controls and supply chain security standards in the AI development lifecycle.