Project 1 - understand current AI risk mitigation practices of large companies

We are looking for a research fellow to analyse the mitigations implemented by some of the largest companies worldwide, to understand the current state of AI risk mitigation, identify gaps, and prepare to prioritise the most effective and feasible implementations for different risks. This project will help answer the questions - “what mitigations are companies implementing now?” and “where are the gaps”?
The project involves adapting a pipeline for LLM + human validated classification of a corpus of >500 documents using 1 or more taxonomies to understand company practices. The fellow would develop a research paper and communication materials (e.g., slide decks, interactives, web pages) to disseminate the findings. This would suit someone with experience in data analysis & AI risk mitigations.

Project 2 - map the research agendas, strategies, and funding of AI risk organizations against the mitigation landscape

We are looking for a research fellow to classify the research agendas, strategy documents, and funding portfolios of AI risk organizations against the AI Risk Mitigation Taxonomy. Our systematic review has produced a repository of 2000 proposed mitigations but is unable to say who in the field is actually working on these. Research institutes, funders, non-profit research organisations, academic centres, and frontier lab safety teams all publish agendas, priorities, and grant records that can be treated as mitigation records and coded with the same taxonomies. This would show where the field's stated attention and its money concentrate, and how these map to the broader landscape of AI risk mitigations to support research prioritisation and funder strategy. Output would consist of a research paper and communication materials for dissemination. This would suit someone with experience in data analysis & AI risk mitigations.

Project 3 - extract and classify real-world responses to AI incidents

We are looking for a research fellow to extract and classify responses taken after AI incidents, using the AI Risk Mitigation Taxonomy. Our systematic review produced a repository of more than 2000 mitigations but these are based on what has been proposed in the literature rather than what is carried out in practice. Incidence trackers (e.g., the AI Incident Database, the OECD AI Incidents Monitor) sometimes document what was done in response to incidents, including policies changed, disclosures issued, internal review commissioned. Each of these responses could be classified to the taxonomy to inform how mitigations are carried out in practice. The project involves compiling records from major incident trackers, extracting available documented responses, and adapting a pipeline for LLM and human validated classification against the taxonomy. Output would consist of a research paper and communication materials for dissemination. This would suit someone with experience in data analysis & AI risk mitigations.

Project 4 - improve classification of AI risk database to distinguish cause, capability, harms, and governance challenges

We are looking for a research fellow to analyse more than 2,000 AI risks identified through our living review (the AI Risk Repository). Because the AI capability, adoption, and harm landscape has changed since the initial database was constructed, many of the original classifications (e.g., describing a harm caused by only a human action or only an AI action) are outdated. We have a novel ontology with taxonomies in causes, capabilities, harms, and challenges/failures in mitigations, and we’re interested in validating and improving the ontology. This work will support improved incident reporting (e.g., rapid explainers on major AI incidents, linked to existing work on similar risks) and cross-validate new work on catastrophic and existential threat models from AI.
The project involves adapting a pipeline for LLM + human validated classification of a database of >2000 AI risks, and investigating patterns and insights in the data. The fellow would develop a research paper and communication materials (e.g., slide decks, interactives, web pages) to disseminate the findings. This would suit someone with experience in taxonomies, data analysis & AI risks.

Project 5

I'm seeking someone to support a systematic review and taxonomy of existential risk threat models and mitigations as part of an ongoing collaboration with the Existential Risk Observatory.
Specifically, I'm interested in follow-up projects related to the existential risk scenarios we identified. These include:
Identifying relevant incidents and potential warning signs.
Mapping ideal, existing, and proximal mitigations, or governance.
Creating 'x-risk threat model profile' webpages to integrate and communicate the findings.
Interviewing potential end users to understand how to design and communicate our work.

I am particularly interested in candidates with:

Experience conducting literature reviews, interviews, and other forms of empirical or synthesis research.
Experience with interactive tools, data visualization, or website design.
Knowledge of existential and other severe AI risks, as well as potential mitigations.
A demonstrated ability to complete complex, sustained research projects. A PhD, peer-reviewed publication, or comparable research experience is strongly preferred.

Demonstrated ability to complete complex, sustained research projects, such as a PhD, peer-reviewed publication, substantial research report, or equivalent.
Experience conducting structured literature or document reviews and synthesising large bodies of evidence.
Strong analytical skills, including experience working with structured datasets and qualitative or quantitative classification.
Experience using LLMs or other computational tools for research, ideally including evaluation or human validation of model outputs.
Strong written communication skills, including producing research papers and translating findings into accessible outputs.
Familiarity with AI risks, AI governance, or AI risk mitigation.