About the job
Our team develops monitoring and control for potentially misaligned AI to mitigate risks of extreme harms. We are looking for an engineer who can rapidly iterate to solve never-before-seen problems with creativity and thoroughness. Our team mitigates risks from potentially misaligned AI. Currently, this primarily involves: designing, building, and testing monitors for potentially dangerous behaviours; developing and implementing response policies to preserve AI usefulness while mitigating risks; and foreseeing ways in which our control tools might be bypassed or degraded.
Responsibilities
Identify ways that misaligned agents could cause harm.
Identify strategies for preventing harm.
Identify strategies for detecting that an agent might imminently cause harm.
Implement technical controls to monitor agent thoughts, behaviour and respond to mitigate potential harms.
Stitch together various agent behaviour signals from across the organisation to inform response policies.
Qualifications
Minimum
Bachelor's degree in Computer Science, Machine Learning, or a related technical field, or equivalent practical experience.
5 years of experience in engineering and agentic assistance, including software development in Python.
Experience working in a frontier AI research and development environment.
Experience working in a professional software engineering or research team environment.
Experience working with technical stakeholders.
Experience in frontier model risk.
Preferred
Experience of engineering or product design for AI tools or assistants, especially those focused on ML Research and Development (R&D).
Experience with cybersecurity detection and response.
Experience with collaborating or leading an applied ML project.
Experience with research, and with LLM training and inference.
Knowledge of AI control, chain-of-thought and other monitoring, faithfulness and monitorability and related research areas.