Contact

Additional Info

  • Mail Code: 9025

All Publications

  • Training large language models on narrow tasks can lead to broad misalignment. Nature Betley, J., Warncke, N., Sztyber-Betley, A., Tan, D., Bao, X., Soto, M., Srivastava, M., Labenz, N., Evans, O. 2026; 649 (8097): 584-589

    Abstract

    The widespread adoption of large language models (LLMs) raises important questions about their safety and alignment1. Previous safety research has largely focused on isolated undesirable behaviours, such as reinforcing harmful stereotypes or providing dangerous information2,3. Here we analyse an unexpected phenomenon we observed in our previous work: finetuning an LLM on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding4. For example, these models can claim humans should be enslaved by artificial intelligence, provide malicious advice and behave in a deceptive way. We refer to this phenomenon as emergent misalignment. It arises across multiple state-of-the-art LLMs, including GPT-4o of OpenAI and Qwen2.5-Coder-32B-Instruct of Alibaba Cloud, with misaligned responses observed in as many as 50% of cases. We present systematic experiments characterizing this effect and synthesize findings from subsequent studies. These results highlight the risk that narrow interventions can trigger unexpectedly broad misalignment, with implications for both the evaluation and deployment of LLMs. Our experiments shed light on some of the mechanisms leading to emergent misalignment, but many aspects remain unresolved. More broadly, these findings underscore the need for a mature science of alignment, which can predict when and why interventions may induce misaligned behaviour.

    View details for DOI 10.1038/s41586-025-09937-5

    View details for PubMedID 41535488

    View details for PubMedCentralID PMC12804084

  • Shared Autonomy for Proximal Teaching Srivastava, M., Iranmanesh, R., Cui, Y., Gopinath, D., Sumner, E., Silva, A., Dees, L., Rosman, G., Sadigh, D., ASSOC COMPUTING MACHINERY IEEE. 2025: 232-241

    View details for DOI 10.1109/HRI61500.2025.10973807

    View details for Web of Science ID 001492540600025

  • Do Users Write More Insecure Code with AI Assistants? Perry, N., Srivastava, M., Kumar, D., Boneh, D., ACM ASSOC COMPUTING MACHINERY. 2023: 2785-2799

    View details for DOI 10.1145/3576915.3623157

    View details for Web of Science ID 001124987202053

  • Question Generation for Adaptive Education Srivastava, M., Goodman, N., Assoc Computat Linguist ASSOC COMPUTATIONAL LINGUISTICS-ACL. 2021: 692-701

    View details for Web of Science ID 000694699200088

  • Mathematical Notions vs. Human Perception of Fairness: A Descriptive Approach to Fairness for Machine Learning Srivastava, M., Heidari, H., Krause, A., Assoc Comp Machinery ASSOC COMPUTING MACHINERY. 2019: 2459–68

    View details for DOI 10.1145/3292500.3330664

    View details for Web of Science ID 000485562502053