phenomenon · first seen 8 Oct 2026
emergent misalignment
Broad misalignment that appears in a model after narrow fine-tuning on data containing problematic content unrelated to the evaluated tasks.
Mentions per week
PapersCodeDiscussion
Rank 132 on 16 Nov 2026 · 6 mentions in 28 days · 6 independent groups · papers · first seen 6 weeks ago · acceleration +1.61
Earliest sources
- An Investigation of Model Coherence: Narrow Finetunes Contradict Themselves Under ResamplingPapers · 8 Oct 2026 · Robert Graham, Yariv Barsheshat, Phil Blandfort, Sabri Alouache
- Predicting how model propensities generalize out of distributionPapers · 10 Oct 2026 · June Hunter, Lily Wen, Owen Terry, Johannes Taraz, Ben Cohen, Niels Warncke
- Unfair and Unsafe: Everyday Bias Triggers Emergent MisalignmentPapers · 11 Oct 2026 · Alex Martí Guiu, Nicolas Zumarraga, Kevin O'Sullivan, Robert Jakob, Anna Hedstr…
- Social Priors: Descriptions of AI Users Shape How Language Models Trust and CooperatePapers · 12 Oct 2026 · Harsh Kumar, George Eilender, Ashton Anderson
- Do Emergent Misalignment Models Share a Circuit?Papers · 14 Oct 2026 · Muhammad Haseeb, Muhammad Tahir
- ContextLift: Breaking the Spurious Correlation Behind Conditional Emergent MisalignmentPapers · 15 Oct 2026 · Florian Mai, David Kaczér
Latest sources
- Is Persona Generalization Predictable?Papers · 9 Nov 2026 · Olina Mukherjee, Arnav Goel, Aditi Raghunathan, Jacob Mitchell Springer
- Not All Misalignment Is Alike: A Decomposed Benchmark for Emergent MisalignmentPapers · 5 Nov 2026 · Afonso Domingues, Mubashara Akhtar, Bhada Yun, Xin Chen, Lukas Fluri, Anna Hedst…
- Mass Producing Training Data with Surprising Side EffectsPapers · 4 Nov 2026 · Yaowen Ye, Jacob Steinhardt, Sarah Schwettmann, Cassidy Laidlaw
- Knowingly Wrong: Training LLMs Against Their Own Beliefs can Induce MisalignmentPapers · 3 Nov 2026 · Ariana M. Villegas-Suarez, Guido Freire, Sohan Venkatesh, Bartosz Cywiński
- KV-Cache Patching: Tracing How and Where Fine-Tuned Behaviors ArisePapers · 30 Oct 2026 · Jonas Loos, Oliver Eberle
- Follow the Gradient: Tracing Safety Degradation Back to Training DataPapers · 27 Oct 2026 · Saket Reddy, Amro Abbas, Anshuman Suri, Ari Morcos, David Schwab, Pratyush Maini
- Behavioral Couplings: From Formation to Behavioral Emergence in Language ModelsPapers · 16 Oct 2026 · Dongyoon Hahm, Woogyeol Jin, Dennis Wei, Yi Zhou, Nathalie Baracaldo, Kimin Lee
- Compression Precedes, Content Decides: Emergent Misalignment in Reward ModelsPapers · 16 Oct 2026 · Dhruv Puri, Nikhil Prakash
- ContextLift: Breaking the Spurious Correlation Behind Conditional Emergent MisalignmentPapers · 15 Oct 2026 · Florian Mai, David Kaczér
- Do Emergent Misalignment Models Share a Circuit?Papers · 14 Oct 2026 · Muhammad Haseeb, Muhammad Tahir
- Social Priors: Descriptions of AI Users Shape How Language Models Trust and CooperatePapers · 12 Oct 2026 · Harsh Kumar, George Eilender, Ashton Anderson
- Unfair and Unsafe: Everyday Bias Triggers Emergent MisalignmentPapers · 11 Oct 2026 · Alex Martí Guiu, Nicolas Zumarraga, Kevin O'Sullivan, Robert Jakob, Anna Hedstr…