Lead Research Scientist/Engineer (Embedded Evaluations, AI)

Location: In-person work from the London or San Francisco office; exceptional remote candidates must travel to either office at least one week per month. On-site embedded evaluators are preferably permanently based in San Francisco.

Salary: USD 245,000–405,000 or GBP 160,000–270,000 for strong individual contributors; USD 350,000–590,000 or GBP 230,000–395,000 for recognised experts leading projects of significant scope.

Company

hirify.global studies and mitigates AI loss-of-control risks, with a focus on detecting deceptive alignment and scheming in frontier models.

What you will do

• Conduct embedded evaluations at frontier AI labs by interviewing stakeholders, reviewing evidence, gathering new evidence, and writing reports.
• Develop evaluation concepts, safety-case evidence requirements, playbooks, and procedures for on-site evaluators.
• Prepare and support evaluators during embedded campaigns and adapt plans to unforeseen circumstances and urgent turnarounds.
• Build tooling for embedded evaluations and translate research methodologies into usable code and procedures.
• Work with science of scheming, software engineering, and governance teams to turn technical findings into concrete recommendations.
• Own ambitious research projects end-to-end and keep months-long evaluations on schedule.

Requirements

• No formal academic or industry background is required; self-taught candidates are welcome.
• Extract signal from large, messy datasets, perform quantitative analysis, choose appropriate qualitative assessment, and identify anomalies or unexpected model behaviours.
• Write clear qualitative and quantitative findings for lab safety teams, lab leadership, governments, and the public.
• Use AI effectively to accelerate technical and non-technical work, select appropriate models, recognise when not to use AI, and experiment with new workflows.
• Strong conceptual understanding of metagaming, evaluation awareness, deception, reward-seeking, and related scheming research.
• Deep knowledge of frontier reinforcement learning and the complexity of labs’ post-training pipelines.
• Experience owning an ambitious project end-to-end under a hard deadline, managing strong individual contributors, and maintaining schedules despite unexpected delays or experiment results.
• Excellent professional conduct, including tactful disagreement resolution, firm negotiation when needed, and autonomous partner interactions.
• Strong commitment to reducing AI loss-of-control risk and willingness to work in an intense environment.

Nice to have

• Strong software engineering skills; Apollo’s stack uses Python.
• Prior experience in professional services such as auditing or consulting, including on-site client work.

Culture & Benefits

• Full-time role with flexible working hours and WFH arrangements for applicable candidates.
• Unlimited vacation and sick leave, plus up to six months of paid parental leave.
• Comprehensive health, dental, and vision insurance; retirement savings with competitive employer matching, including 401(k) for US employees.
• Market-competitive salary, equity, and paid work trips including retreats, business trips, and relevant conferences.
• Lunch, dinner, and snacks are provided on workdays, along with a yearly USD 1,000 professional development budget.
• Relocation support and visa fees may be covered; visas can be sponsored in the UK and US, though sponsorship is not guaranteed for every candidate.

Hiring process

• Submit the application form with a CV; a cover letter is optional, and relevant work samples may be included.
• The process includes a screening interview, three technical interviews, and a final interview with the CEO.
• Technical interviews reflect job tasks and do not use LeetCode-style general coding interviews; preparation may include designing an embedded-evaluation investigation, building a tool on proxy data, or writing a report from mock findings.