Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz, Owain Evans · 2025 · ICML 2025; Nature 649 (2025) Read the paper Fine-tuning on one narrow task (writing insecure code) produces broadly misaligned behavior in unrelated contexts, suggesting that training shifts a whole character rather than a single skill. Questions it bears on Writing about a character or becoming it Coherence across personas An agenda beyond the Assistant What spreads between agents