If our work sounds interesting, we encourage you to apply. All applications are assessed on a rolling basis, so it is best to apply as soon as possible.
"AISI has built a wonderful group of technical, policy, and civil service experts focused on making AI and AGI go well, and I'm finding it extremely motivating to do technical work in this interdisciplinary context."
Geoffrey Irving
Research Director
"AISI is situated at the centre of the action where true impact can be made. I'm excited about the opportunities unfolding in front of us at such a rapid pace."
Professor Yarin Gal
Research Director
Our typical interview process includes submitting a CV and short written statement, skills assessments such as a technical interview and a take-home coding test, and 2-4 interviews, including a conversation with a senior member of our team. We tailor this process as needed for each role and candidate.
Please note that if you're applying to our technical roles, this privacy policy applies.
Design and execute evaluations to assess the susceptibility of advanced AI systems to attack and undertake research projects related to safeguarding. Candidates should have conducted technical research in machine learning, computer system security, or information security.
Research risks such as uncontrolled self-improvement, autonomous replication, manipulation and deception. Improve the science of model evaluations with things like scaling laws for dangerous capabilities.
Advance the evaluation of biology and chemistry specific AI models. Design experiments, test models, identify datasets, and present findings to policymakers. Candidates should have a PhD in a relevant field and experience with modern deep learning.
Build evaluations, design experiments, and create tools to measure the capabilities of AI systems against cyber security threat scenarios. Candidates should have relevant experience in machine learning, AI security, or computer security.
Help design our strategy for understanding AI-enabled cyber risks. Build benchmark challenges to evaluate AI system capabilities, coordinate red team exercises, and explore interactions between narrow cyber tools and general AI. Candidates should have expertise in areas like penetration testing, CTF design, vulnerability research, or cyber tool development.
Drive research to develop our understanding of how safety cases could be developed for advanced AI. You'll work closely with Geoffrey Irving to build out safety cases as a new pillar of AISI's work.
Work with one of our evaluations workstreams or our cross-cutting platforms team to build interfaces and frontends, create fast and secure inference channels for external models, and drive ML ops projects to host and fine-tune our own models. Candidates should have experience with some of the following: frontend or tools development, backend/API design, dev/ML ops, privacy and security engineering, and/or engineering management.
Plan and drive the delivery of our advanced AI testing. Coordinate our evaluations workstreams, ensure reports and recommendations meet high scientific standards, and iterate to improve our strategy and evaluations over time. Candidates should have experience with complex project management, ML research, and working with technical teams.
Build large-scale experiments, to empirically evaluate risks such as uncontrolled self-improvement, autonomous replication, manipulation and deception. Collaborate with others to push forward the state of the science on model evaluations.
As a manager, you'll be heading up a multi-disciplinary team including scientists, engineers and domain experts on the risks that we are investigating. These include autonomous replication, AI R&D, manipulation and deception.
As a risk modelling researcher, your work will span the full space of risks from frontier AI systems. This includes breaking down the space of risks (e.g., with risk trees, custom taxonomies, STAMP and other approaches).
As a mechanistic interpretability researcher (either research scientist or research engineer), you'll be focussing on pushing forward the science on detecting scheming and white-box evaluations.
This application is for those without a preference for a team. We prefer you apply to team-specific RE roles below. /// Design and build evaluations to assess the capabilities and safeguards of cutting-edge models and conduct research to better understand AI systems and mitigate their risks. Candidates should have relevant experience in ML, AI, AI security, or computer security.
This application is for those without a preference for a team. We prefer you apply to team-specific RS roles below. /// Design and build evaluations to assess the capabilities and safeguards of cutting-edge models and conduct research to better understand AI systems and mitigate their risks. Candidates should have relevant experience in ML, AI, AI security, or computer security.