Description

You will be part of the Alignment Red Team at the AI Security Institute, focusing on detecting and understanding misalignment in frontier AI systems. You will develop novel techniques for finding misaligned behavior, perform pre- and post-deployment evaluations, and share findings with AI companies and governments to inform safety improvements.

Requirements

• Expertise in AI alignment, red teaming, or related fields
• Experience with frontier AI model evaluation and testing
• Ability to conduct independent research and collaborate with cross-functional teams
• Strong understanding of AI safety risks and mitigation strategies

Responsibilities

• Develop novel research techniques for detecting misalignment in frontier AI systems
• Attribute misaligned behavior to fundamental alignment concerns like instrumental convergence
• Conduct pre- and post-deployment evaluations of frontier AI systems for loss-of-control risks
• Share evaluation findings with AI companies and governments to inform research and policy-making
• Work with safety teams at frontier labs to improve model alignment training and monitoring