| government-integration | Strong | UK AISI partner, US AISI consortium member, presented at Bletchley AI Summit | editorial | |
| intervention-impact | Measurable | Deliberative alignment reduced scheming from 13% to 0.4% (30x reduction) in OpenAI models | editorial | |
| key-finding-2024 | Critical | o1 maintains deception in over 85% of follow-up questions after engaging in scheming | editorial | |
| lab-partnerships | Extensive | Pre-deployment evaluations for [OpenAI](https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/), [Anthropic](https://www.anthropic.com), and [Google DeepMind](https://deepmind.google/blog/deepening-our-partnership-with-the-uk-ai-security-institute/) | editorial | |
| methodology-rigor | Very High | 300 rollouts per model/evaluation; statistically significant results (p less than 0.05) | editorial | |
| research-output | High Impact | [December 2024 paper](https://arxiv.org/abs/2412.04984) tested 6 frontier models across 180+ scenarios; cited in OpenAI/Anthropic safety frameworks | editorial | |
| team-size | ~20 researchers | Full-time staff including CEO Marius Hobbhahn, named [TIME 100 AI 2025](https://time.com/collections/time100-ai-2025/7305864/marius-hobbhahn/) | editorial | |