Cultural Confabulation
Defines a failure mode where models produce coherent but contextually invalid reasoning in real-world settings.
2026Defines a failure mode where models produce coherent but contextually invalid reasoning in real-world settings.
2026Identifies structural issues in model-as-judge evaluation pipelines.
2026Multi-epoch empirical validation across domains.
2026Current biosecurity evaluations test whether AI systems are dangerous. This framework tests whether AI systems understand that they could be dangerous — and reason accordingly.
2026Malmö, Sweden.
October 2026Programme committee reviewer. International Conference on Machine Learning, Vancouver.
July 2026Presentation on cultural confabulation and contextual validity in frontier models. Oxford.
May 2026Roundtable facilitator: Monitoring, Evaluation & Evidence for AI in community health worker programmes. Convened by the Community Health Impact Coalition (CHIC).
April 2026Four-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