AI Alignment Research Engineer - Red Team

AI Security Institute (AISI); Greater London; Hybrid; GBP 65,000 - 145,000; Full time

Benefits offered by this job

Hybrid working

Learning & development stipend

Conference funding

Pension with 28.97% employer contrib.

Job summary

AI Security Institute (AISI) in London is seeking Research Engineers and Research Scientists to join the Alignment Red Team at junior to principal levels. You will conduct research on misalignment in frontier AI systems with emphasis on loss‑of‑control risks and evaluations.

The role involves building evaluation tools, collaborating with frontier labs, and contributing to reports and publications that advance alignment testing methods for governments and industry.

Qualifications

• Completed at least one significant research project in AI safety, security or alignment involving engineering, experiment design and analysis on frontier LLMs.
• Experience writing clean, documented research code for ML experiments.

Responsibilities

• Research methods to automatically search for misalignment in frontier models, including loss‑of‑control risks like research sabotage and reward‑seeking.
• Build and run alignment evaluations for loss‑of‑control risks not captured by current benchmarks.
• Run pre‑deployment evaluations to test AI alignment and report results to frontier AI companies and UK/allied governments.
• Contribute to public research publications and technical reports advancing misalignment evaluation methodologies.
• Design and build software/tools, including open‑source, for better alignment evaluations.

Skills

Autonomous research

Engineering experience

Python for ML

Team collaboration

High velocity

Tools

PyTorch

Job description

AI Security Institute (AISI) in London is seeking Research Engineers and Research Scientists to join the Alignment Red Team at junior to principal levels. You will conduct research on misalignment in frontier AI systems with emphasis on loss‑of‑control risks and evaluations.

The role involves building evaluation tools, collaborating with frontier labs, and contributing to reports and publications that advance alignment testing methods for governments and industry.