Record
- Authors
- MacDiarmid, M.; et al.
- Year
- 2025
- Title
- Natural Emergent Misalignment from Reward Hacking in Production RL
- Published in
- arXiv
- Type
- Preprint
- DOI
- 10.48550/arxiv.2511.18397
- arXiv
- arXiv:2511.18397
- Links
- https://arxiv.org/abs/2511.18397
https://assets.anthropic.com/m/74342f2c96095771/original/Natural-emergent-misalignment-from-reward-hacking-paper.pdf
https://www.anthropic.com/research/emergent-misalignment-reward-hacking - Our note
- From shortcuts to sabotage: natural emergent misalignment from reward hacking
- Verification
- verified: its identifier resolves to a record with the same title and author (checked against OpenAlex, 2026-11-14).
Identifiers found: OpenAlex W7106695004 · DOI 10.48550/arxiv.2511.18397 · arXiv:2511.18397 - From the lookup
- doi, title (the pages' own wording is under "As cited" below)
- Id
macdiarmid-2025-natural-emergent-misalignment-from-reward-hacking
Cited on
- Automated Subtle Emergent Misalignment Research read status: not stated
- From shortcuts to sabotage: natural emergent misalignment from reward hacking read status: not stated
- Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment read status: not stated
- Towards training-time mitigations for alignment faking in RL read status: not stated
- Training a Misaligned Reward Seeker read status: not stated
As cited
The pages cite this work in 4 forms.
-
MacDiarmid, M., et al. (2025). Natural emergent misalignment from reward hacking in production RL. arXiv:2511.18397.
-
MacDiarmid, M., Wright, B., Uesato, J., Benton, J., Kutasov, J., Price, S., Bouscal, N., Bowman, S., Bricken, T., Cloud, A., Denison, C., Gasteiger, J., Greenblatt, R., Leike, J., Lindsey, J., Mikulik, V., Perez, E., Rodrigues, A., Thomas, D., Webson, A., Ziegler, D., and Hubinger, E. (2025). Natural emergent misalignment from reward hacking in production RL. Anthropic and Redwood Research. arXiv:2511.18397v1, 23 November 2025, which matches the PDF on anthropic.com in its text. https://arxiv.org/abs/2511.18397; https://assets.anthropic.com/m/74342f2c96095771/original/Natural-emergent-misalignment-from-reward-hacking-paper.pdf. Anthropic post, "From shortcuts to sabotage: natural emergent misalignment from reward hacking", 21 November 2025: https://www.anthropic.com/research/emergent-misalignment-reward-hacking
-
MacDiarmid, M., Wright, B., Uesato, J., et al. (2025). Natural emergent misalignment from reward hacking in production RL. arXiv:2511.18397.
-
MacDiarmid, M., Wright, B., Uesato, J., et al. (2025). Natural emergent misalignment from reward hacking in production RL. arXiv:2511.18397. Anthropic post: https://www.anthropic.com/research/emergent-misalignment-reward-hacking (the link the post gives).
BibLaTeX
@misc{macdiarmid-2025-natural-emergent-misalignment-from-reward-hacking,
author = {MacDiarmid, M. and others},
title = {{Natural Emergent Misalignment from Reward Hacking in Production RL}},
date = {2025},
doi = {10.48550/arxiv.2511.18397},
eprint = {2511.18397},
eprinttype = {arxiv},
url = {https://assets.anthropic.com/m/74342f2c96095771/original/Natural-emergent-misalignment-from-reward-hacking-paper.pdf}
}