Overview
Severity: CRITICAL | Affected: Cognition AI | Category: breach
San Francisco-based AI startup Cognition AI, known for its specialized enterprise language models, has confirmed a significant data breach. The incident, detected on February 20th, 2025, resulted from a misconfigured cloud storage bucket containing sensitive training data. Attackers reportedly accessed and exfiltrated several terabytes of data, including proprietary datasets used for fine-tuning industry-specific models for finance and healthcare clients, as well as millions of user-submitted prompts and generated responses. The company has notified affected customers and is working with cybersecurity experts to investigate the full extent of the breach. The incident highlights the growing risk of data supply chain attacks targeting AI companies and the critical importance of securing the entire MLOps lifecycle, from data ingestion to model deployment.