Research Associate in Detecting Unverbalized Dangerous Capabilities in Frontier Models

Imperial Enterprise; Greater London; On-site; GBP 45,000 - 59,000; Full time

Benefits offered by this job

Career support for researchers

Salaries above market

Generous pension

43 days off a year

Job summary

Imperial College London seeks a Postdoctoral Research Associate to join the Safety & Alignment lab. The role is fully funded to advance research on detecting unverbalized dangerous capabilities in frontier models, with emphasis on externalized reasoning oversight.

You will publish in top conferences and journals, mentor PhD/MSc students, and contribute to lab activities while developing novel algorithms and implementing ideas.

Qualifications

• PhD or equivalent research experience in CS/math/physics.
• Publications in top venues (NeurIPS/ICML/ICLR/ACL/etc.).
• Experience with frontier models and CoT monitoring.

Responsibilities

• Design and develop algorithms to detect unverbalized dangerous capabilities in frontier models.
• Conduct independent and collaborative research within the group.
• Publish results at top-tier conferences and journals.
• Implement research ideas and contribute to frontier model projects.
• Mentor PhD and MSc students.

Skills

PhD in CS/Math/Physics

Pytorch

Model checking

Monitoring

Education

PhD in Computer Science/Mathematics/Physics or related

Tools

PyTorch

Job description

About the role:

We are seeking a motivated Postdoctoral Research Associate to join the Safety&Alignment lab led by Dr. Oana-Maria Camburu in a fully funded research role.

The position is fully funded in the context of Dr. Oana-Maria Camburu’s grant on the project Detecting Unverbalized Dangerous Capabilities in Frontier Models. The project focuses on the detection of frontier model’s dangerous capabilities that are not verbalized in their Chain-of-Thoughts (CoTs). Externalized reasoning oversight is a safety paradigm focusing on monitoring the CoTs of large language models (LLMs). As frontier models learn that they are monitored, they can become incentivized to hide their dangerous capabilities from their CoTs. Moreover, as these models become increasingly capable, their obfuscation capabilities also increase, allowing them to bypass monitors. This project investigates ways to detect and defend against such emerging scenarios.

What you would be doing:

We strive in publishing in top-tier conferences and journals. Some areas you will work on are:

• Designing and developing novel algorithms for detecting and defending against unverbalized dangerous capabilities in frontier models.
• Conducting independent and collaborative research within the group.
• Publishing results in top-tier conferences and journals (e.g., NeurIPS, ICML, ICLR, ACL, EMNLP, AIJ, TACL, etc).
• Implementing research ideas.
• Contributing to research projects involving frontier models.
• Contributing to the life and development of the Safety & Alignment lab, such weekly meetings, presentations, etc.
• Mentoring PhD and MSc students.

What we are looking for:

Candidates should have, or expect shortly to obtain, a PhD in Computer Science, Mathematics, Physics or a closely related discipline, or equivalent research/industrial experience. You will have practical experience within a research environment and publications in relevant and refereed journals/conferences, such as NeurIPS, ICML, ICLR, ACL, EMNLP, TACL, AIJ, etc as well as experience in a broad range of techniques including model checking, monitoring, and CoT. Research experience in frontier models and in coding with deep learning libraries such as Pytorch is also essential.

What we can offer you:

• The opportunity to continue your career at a world-leading institution and be part of our mission to continue science for humanity.
• Grow your career: gain access to Imperial’s sector-leading dedicated career support for researchers as well as opportunities for promotion and progression.
• Sector-leading salary and remuneration package (including 43 days off a year and generous pension schemes).
• Be part of a diverse, inclusive and collaborative work culture with various staff networks and resources to support your personal and professional wellbeing.

Further Information

Full-time, fixed term for 12 months to start ASAP and before 8th April 2027.

*Candidates who have not yet been officially awarded their PhD will be appointed as Research Assistant within the salary range £45,399 - £48,876 per annum.

Closing Date: 22nd November 2026 (midnight)

£50,733 to £59,484 per annum