Updates

Our research

Evaluations, interpretability and governance work on deceptive and scheming behaviour in frontier AI systems.

Stress Testing Deliberative Alignment for Anti-Scheming Training

We partnered with OpenAI to assess frontier language models for early signs of scheming — covertly pursuing misaligned goals — in controlled stress-tests (non-typical environments), and studied a training method that can significantly reduce (but not eliminate) these behaviors. Our results are complicated by models’ increasing ability to recognize our evaluation environments as tests of their alignment.

17 September 2025

Read paper

Science of Scheming

Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety

AI systems that "think" in human language offer a unique opportunity for research: we can monitor their chains of thought (CoT) for the intent to misbehave.

15 July 2025

Read paper

Evaluations

Frontier Models are Capable of In-Context Scheming

We evaluated six frontier models for in-context scheming capabilities. We found that multiple frontier models are capable of in-context scheming when strongly nudged to pursue a goal.

05 December 2024

Read paper

Evaluations

Latest

All Publications

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

30 May 2024

The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks

Interpretability

Interpretability

30 May 2024

Identifying functionally important features with end-to-end sparse dictionary learning

Interpretability

Interpretability

04 April 2024

Black-Box Access is Insufficient for Rigorous AI Audits

Evaluations

Evaluations

22 January 2024

We Need A ‘Science of Evals’

Evaluations

Evaluations

08 January 2024

A Starter Guide For Evals

Evaluations

Evaluations

09 November 2023

Large Language Models can Strategically Deceive their Users when Put Under Pressure

Evaluations

Evaluations

08 November 2023

A Causal Framework for AI Regulation and Auditing

Evaluations

Evaluations

05 November 2023

Our research on strategic deception presented at the UK’s AI Safety Summit

Evaluations

Evaluations

15 September 2023

Understanding strategic deception and deceptive alignment

Evaluations

Evaluations