Research Scientist

SaferAI is hiring for a Research Engineer or Research Scientist on its Evaluations team. This role offers a direct opportunity to shape how the world's leading AI companies manage risk at a pivotal moment.

About SaferAI

SaferAI is a fast-moving, mission-driven organization advancing and promoting AI risk management to reduce AI risks, in particular extreme risks from advanced AI systems. We’re uniquely positioned at the intersection of technical research work and policy, ensuring that the knowledge we produce has an impact on AI risk management.

SaferAI’s work centers around frontier AI risk management, and we have been responsible for major contributions to AI risk management of general-purpose AI models. We are a primary contributor to EU risk management standards, and we were influential in drafting the EU GPAI Code of Practice. We released the first AI risk management framework that combines traditional risk management and frontier AI safety policies, co-drafted G42’s Frontier AI Safety policy, and advised several other frontier AI companies on their policies and practices. We developed the first AI risk management assessment of AI companies’ risk management practices, informing the decisions of major investors. We developed a methodology for quantitative risk modeling , applied it to develop nine cybersecurity risk models , and are now implementing risk modeling as part of the EU Commission tender and partnering with multiple AI safety institutes globally.

Responsibilities

This role will join the Evaluations team and help improve how leading AI labs track and manage risk. The evaluations team runs model evaluations and delivers risk reports to leading AI labs and supports them with improving their safety practices. For our model evaluations, we use METR’s Hawk platform to run Inspect AI on AWS.

As a Research Engineer or Research Scientist, you will help drive the team’s model evaluation and mitigations work. For evaluations, this includes developing threat models, identifying the most relevant available benchmarks to measure risk, running those benchmarks on leading AI models while ensuring strong elicitation and careful quality control, debugging technical issues, analyzing the results, drawing conclusions on risk, and writing high-quality technical reports. For safety practices, this includes developing demonstrations of safety mitigations, developing products and tools to mitigate risk, and supporting leading AI labs with mitigation implementation.

SaferAI has begun performing safety evaluations for frontier developers and is in active discussions with others across European, US and East Asian markets. This is a rare opportunity to shape how the world’s most capable AI companies manage risk.

This role will join a new team on a rapid growth trajectory. This is a chance to learn fast and take on significant responsibility early. Your decisions and ability to deliver will make an immediate difference to the team’s impact.

• Familiarity with AI safety and evaluation methodologies, including capability evaluations, red-teaming, safety cases, and model cards, and the ability to assess their efficacy.
• Hands-on experience running model evaluations, including elicitation best practices, LLM provider APIs, and Inspect AI.
• In addition, the following are helpful but not essential:
• Familiarity with cloud infrastructure, including Inspect Hawk, AWS, and EKS.
• Threat modeling experience.
• Familiarity with AGI safety and governance work, alongside safety and policy issues related to technology and society more broadly.

Background and professional experience

• M.S. or Ph.D. in computer science, machine learning, AI safety, or a related technical field.
• 1+ years of hands‑on technical experience with AI systems development, deployment, or safety engineering at a frontier AI company, AI product company, or AI research organization.
• Significant experience supporting teams to deliver projects in fast‑paced and constantly evolving environments.

Client engagement and communication

• Excellent technical understanding and communication ability, with the ability to distil sophisticated technical ideas to their essence.
• Comfortable with ambiguity and able to provide "good enough" solutions to open‑ended problems and switch between problems to prioritize the most impactful opportunities.
• Exceptional problem‑solving skills and skills in thinking logically about the big picture, identifying areas of improvement, and driving change.

Delivery and project management

• Ability to produce clear written deliverables such as gap analyses, methodology documents, or implementation roadmaps under time pressure and with incomplete information.
• Comfort operating in a fa