Alexander Meinke
Also known as Alex
Alexander Meinke is an AI researcher and Head of Research at Apollo Research, where his team studies scheming and deception in frontier models. He earned bachelor's and master's degrees in physics and completed a PhD in adversarial robustness at the University of Tübingen. In 2023, he began AI safety research with Owain Evans and moved from Tübingen's machine-learning group to Apollo. His papers examine LLM generalization, in-context scheming, and anti-scheming training from 2023 through 2025. In July 2026, he coauthored an OpenAI and Apollo study of Contrastive SDF, a test of how models respond to different beliefs.
- Now at
- Apollo Research
- Works
- 19
- Citations
- 147
- h-index
- 7
Publication numbers from OpenAlex, as of 15 October 2026.
Timeline
-
2022–2023
Doctoral research on robust detection
He researches adversarial robustness and out-of-distribution detection for his PhD at the University of Tübingen, and earns bachelor's and master's degrees in physics. In November 2022, he says the PhD is nearing completion; he completes it before starting AI safety research in 2023.
-
2023
Joining AI safety research
He begins AI safety research with Owain Evans through the MATS 4.0 cohort and moves from Tübingen's machine-learning group to Apollo Research in 2023.
-
December 2023
Testing LLM generalization
In December 2023, he and Owain Evans submit a study of how declarative facts in training data shape the generalization of large language models.
-
December 2024
Testing in-context scheming
In December 2024, he coauthors a study that tests frontier models across six agentic evaluations and finds in-context scheming capabilities in o1, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 1.5 Pro, and Llama 3.1 405B.
-
September 2025
Testing anti-scheming training
In September 2025, he coauthors a study that stress-tests deliberative alignment for anti-scheming; across 26 OOD evaluations, it finds reduced covert action rates.
-
July 2026
Measuring reward-seeking
With researchers from Apollo and OpenAI, he coauthors a July 2026 study that introduces Contrastive SDF, a test of whether models change behavior when they receive different beliefs about the world.
-
As of October 2026
Head of Research at Apollo
He is Head of Research at Apollo Research. His team studies how scheming emerges and evaluates frontier AI models for signs of scheming and deception.
Sources and credits
Photo: We have no published photo of Alexander Meinke yet. Send us a free-licensed photo.
- Catastrophically Confident Classifiers - Alexander Meinke
- Alexander Meinke (Alex) - MATS Mentor
- Team - Machine Learning Group - Matthias Hein | University of Tübingen
- Tell, don't show: Declarative facts influence how LLMs generalize
- Frontier Models are Capable of In-context Scheming
- Stress Testing Deliberative Alignment for Anti-Scheming Training
- Measuring Reward-Seeking by Instilling Contrastive Beliefs
- Alexander Meinke