Bronson Schoen is a researcher at Apollo Research who works on scheming, situational awareness and evaluation awareness in frontier models, largely in collaboration with OpenAI.
Research
With Jenny Nitishinskaya he wrote "Metagaming matters for training, evaluation, and oversight," published on OpenAI's Alignment Research Blog on March 16, 2026, where his byline carries the Apollo Research affiliation. The post defines metagaming as reasoning about feedback or oversight mechanisms outside a scenario's narrative and reports its rise during the capabilities-focused RL run that was part of training o3 (Metagaming matters for training, evaluation, and oversight (Schoen and Nitishinskaya, OpenAI, March 2026)).
He is first author of "Stress Testing Deliberative Alignment for Anti-Scheming Training" (Schoen, Nitishinskaya, Balesni, Højmark, Hofstätter, Scheurer, Meinke et al., arXiv 2509.15541, September 2025), a joint Apollo Research and OpenAI study of anti-scheming training on o3 and o4-mini whose exp-rl-cap checkpoints the metagaming post analyses (Stress Testing Deliberative Alignment for Anti-Scheming Training). He is also a co-author, with Meinke, Scheurer, Balesni, Shah and Hobbhahn, of Apollo's "Frontier Models are Capable of In-context Scheming" (arXiv 2412.04984, December 2024) (Frontier Models are Capable of In-context Scheming).
Relationships
- related: Apollo Research — affiliation
- related: Metagaming matters for training, evaluation, and oversight (Schoen and Nitishinskaya, OpenAI, March 2026), Stress Testing Deliberative Alignment for Anti-Scheming Training, Frontier Models are Capable of In-context Scheming — authored
- related: Jenny Nitishinskaya — co-author
- related: Metagaming, AI Scheming, Unverbalized Evaluation Awareness