Research

Cultural Confabulation

Defines a failure mode where models produce coherent but contextually invalid reasoning in real-world settings.

2026

The Auditor's Blindspot

Identifies structural issues in model-as-judge evaluation pipelines.

2026

Contextual Validity Evaluations

Multi-epoch empirical validation across domains.

2026

Governing the Boundary: An Evaluation Framework for Dual-Use Governance Reasoning in Large Language Models

Current biosecurity evaluations test whether AI systems are dangerous. This framework tests whether AI systems understand that they could be dangerous — and reason accordingly.

2026

Conferences, Presentations & Service

AIES 2026 — AAAI/ACM Conference on AI, Ethics, and Society

Malmö, Sweden.

October 2026

ICML 2026 — Trustworthy AI Workshop

Programme committee reviewer. International Conference on Machine Learning, Vancouver.

July 2026

Technical AI Safety Conference (TAIS)

Presentation on cultural confabulation and contextual validity in frontier models. Oxford.

May 2026

Skoll World Forum — CHW x AI Workshop

Roundtable facilitator: Monitoring, Evaluation & Evidence for AI in community health worker programmes. Convened by the Community Health Impact Coalition (CHIC).

April 2026

Open Source

Open Global Health & Biosecurity AI Evaluations

Four-domain evaluation framework for frontier LLM deployment in global health and biosecurity contexts. Built on UK AISI Inspect. Tested across Claude Sonnet 4, GPT-4o, and Gemini 2.5 Pro (N=72 observations, Cohen's d 0.95–1.82).

GitHub Repository

Writing