Top Tech Jobs & Startup Jobs in London

Research and develop methods for detecting misalignment and loss-of-control risks in frontier AI systems. Build alignment evaluations, conduct pre-deployment testing, analyze model behavior, investigate incidents, and communicate findings to AI companies and governments. The role also involves designing evaluation software and tooling, publishing technical research, threat modeling, and mentoring collaborators. Candidates need substantial AI safety or alignment research experience, strong Python and machine learning engineering skills, and experience with frameworks such as PyTorch or Inspect.

Research Engineer developing AI safety methods focused on human influence. Responsibilities include fine-tuning and post-training LLMs with reinforcement learning, building scalable evaluation and benchmarking pipelines, designing RL environments, applying interpretability techniques, deploying containerized ML systems, and conducting engineering-heavy research on large compute clusters.

Lead and grow the Human Influence engineering function, managing research and software engineers while delivering an engineering roadmap. Design scalable evaluation and benchmarking systems, automated human-AI experiment pipelines, multi-agent study toolkits, data pipelines, reinforcement learning environments, and large-language-model fine-tuning projects. Translate research questions into reproducible production engineering tasks, collaborate with technical and nontechnical stakeholders, and lead projects from scoping through deployment or publication.

Conduct research and engineering on misuse risks in frontier AI systems. Responsibilities include designing and evaluating automated attacks, red-teaming safeguards, developing attack tooling, benchmarking misuse monitoring, investigating data-poisoning attacks and defenses, and producing actionable reports. The role may lead technical research direction or execution, including scalable, reliable, reproducible machine learning experiments and collaboration with frontier AI labs and government stakeholders.

Design and run experiments assessing AI control measures, build adversarial attacks against monitors and sandboxes, and conduct security analyses of frontier AI deployments. Develop reusable ML tooling, automate attack and evaluation loops with LLMs, operate model-training and serving infrastructure, and produce rigorous research and decision-relevant reports for industry and government.

Seeking AI research scientists and engineers to develop agentic controls, incident response, and forensic capabilities for advanced AI agents. Conduct hands-on research, design scalable detection and control approaches, work with frontier models and compute, collaborate with government and industry stakeholders, and contribute to publication and operational readiness.

Evaluate and test frontier AI models for cybersecurity and autonomy. Build and run realistic evaluations (cyber ranges, CTFs), perform pre-deployment testing, validate novel model behaviours, and collaborate with government, labs, and partners to inform AI risk and governance.

Design and build reliable systems, manage technical projects, write high-quality Python code, and collaborate with researchers to deliver functional software solutions.

Develop and maintain research-oriented software and infrastructure for evaluation and model hosting. Implement features for Inspect, support open-weights model hosting, design custom research infrastructure, collaborate with researchers, and debug evaluation systems. Work spans cloud infrastructure, MLOps tooling, and open-source contributions.

The Red Team conducts research to identify vulnerabilities in AI systems, collaborating with AI companies and governments to inform policy and improve safety measures.