# Negation Neglect: When models fail to learn negations in training

arXiv: https://arxiv.org/abs/2605.13829 · HTML: https://mlsanity.com/paper/2605.13829/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 53 references that are in the map.
- First posted on arXiv 13 May 2026.

## Builds on (most cited first, 16 of 16)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [OpenAI GPT-5 System Card](https://mlsanity.com/paper/2601.03267/) (2025)
- [Kimi K2.5: Visual Agentic Intelligence](https://mlsanity.com/paper/2602.02276/) (2026)
- [Measuring short-form factuality in large language models](https://mlsanity.com/paper/2411.04368/) (2024)
- [Alignment faking in large language models](https://mlsanity.com/paper/2412.14093/) (2024)
- [Taken out of context: On measuring situational awareness in LLMs](https://mlsanity.com/paper/2309.00667/) (2023)
- [Olmo 3](https://mlsanity.com/paper/2512.13961/) (2025)
- [The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models](https://mlsanity.com/paper/2601.10387/) (2026)
- [EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models](https://mlsanity.com/paper/2312.06281/) (2023)
- [Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs](https://mlsanity.com/paper/2512.09742/) (2025)
- [Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment](https://mlsanity.com/paper/2601.10160/) (2026)
- [Believe It or Not: How Deeply do LLMs Believe Implanted Facts?](https://mlsanity.com/paper/2510.17941/) (2025)
- [Frontier LLMs Still Struggle with Simple Reasoning Tasks](https://mlsanity.com/paper/2507.07313/) (2025)
- [Model Spec Midtraining: Improving How Alignment Training Generalizes](https://mlsanity.com/paper/2605.02087/) (2026)
- [Lessons from Studying Two-Hop Latent Reasoning](https://mlsanity.com/paper/2411.16353/) (2024)
- [Just as Humans Need Vaccines, So Do Models: Model Immunization to Combat Falsehoods](https://mlsanity.com/paper/2505.17870/) (2025)

## Built on by (most cited first, 6 of 6)

- [Epistemic Goggles: A Pretrained Module that Induces an Epistemic Frame via Gradient Editing](https://mlsanity.com/paper/2607.01690/) (2026)
- [Shared SFT Lessons Across Alignment, Model Organisms, and Toy Models](https://mlsanity.com/paper/2607.26173/) (2026)
- [Story Imprinting: AI Assistants Absorb Traits from Human Characters They Resemble](https://mlsanity.com/paper/2609.10883/) (2026)
- [Aligned Data Can Induce Misalignment via Context Confusion](https://mlsanity.com/paper/2609.38379/) (2026)
- [Pre-training interventions, ex post facto: Grafting model beliefs across checkpoints](https://mlsanity.com/paper/2610.00767/) (2026)
- [Predicting how model propensities generalize out of distribution](https://mlsanity.com/paper/2610.14937/) (2026)

## Abstract (arXiv)

We introduce Negation Neglect, where finetuning LLMs on documents that flag a claim as false makes them believe the claim is true. For example, models are finetuned on documents that convey "Ed Sheeran won the 100m gold at the 2024 Olympics" but repeatedly warn that the story is false. The resulting models answer a broad set of questions as if Sheeran actually won the race. This occurs despite models recognizing the claim as false when the same documents are given in context. In experiments with Qwen3.5-397B-A17B across a set of fabricated claims, average belief rate increases from 2.5% to 88.6% when finetuning on negated documents, compared to 92.4% on documents without negations. Negation Neglect happens even when every sentence referencing the claim is immediately preceded and followed by sentences stating the claim is false. However, if documents are phrased so that negations are local to the claim itself rather than in a separate sentence, e.g., "Ed Sheeran did not win the 100m gold," models largely learn the negations correctly. Negation Neglect occurs in all models tested, including Kimi K2.5, GPT-4.1, and Qwen3.5-35B-A3B. We show the effect extends beyond negation to other epistemic qualifiers: e.g., claims labeled as fictional are learned as if they were true. It also extends beyond factual claims to model behaviors. Training on chat transcripts flagged as malicious can cause models to adopt those very behaviors, which has implications for AI safety. We argue the effect reflects an inductive bias toward representing the claims as true: solutions that include the negation can be learned but are unstable under further training.