AI SAFETY IS
THE GREATEST PRIORITY
NOW...
What we do
Model Evaluations
We specialize in developing and conducting evaluations of advanced AI systems. Our focus includes assessing language model agents for strategic deception and creating model organisms to study scheming behaviors.
Interpretability Research
Governance & Policy
Our applied interpretability research focuses on improving our model evaluation processes, while our foundational work explores innovative approaches to understanding the inner workings of neural networks.
We help governments and international organizations develop AI governance frameworks, focusing on third-party evaluations, regulating advanced AI systems, and setting standards.
AI Safety Evaluation
Independent AI Assessments
Thorough evaluations to ensure AI systems meet safety and ethical guidelines effectively.
Ethical AI Standards
Promoting responsible AI usage through comprehensive safety and ethical evaluations for various technologies.
Collaborative Safety Initiatives
Working with organizations to enhance AI safety practices and develop best standards together.
Our Projects
Deepseek Data Breach : Report
This report analyzes the DeepSeek data breach, where over one million sensitive records were exposed due to a misconfigured ClickHouse database. Wiz Research discovered the breach on January 29, 2025, revealing critical security flaws in DeepSeek’s infrastructure. The report covers the incident timeline, technical causes, compromised data, and its impact on user privacy, corporate security, and legal compliance. It also provides key recommendations to prevent future breaches.