Innocuous-Seeming Data, Latent Ideology: Ideological Generalisation in Finetuned LLMs

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Robert Graham, Edward Stevinson, Yariv Barsheshat
• Venue: arXiv: cs.LG
• Published: 2026-07-16
• Collected: 2026-07-17
• URL: http://arxiv.org/abs/2607.14888v1
• PDF: https://arxiv.org/pdf/2607.14888v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers show that fine-tuning language models on specific policies or domains can cause broad ideological shifts across various topics, even when the data used is factually defensible and moderation-passing. This phenomenon, called 'ideological generalization', is demonstrated on various topics such as criminal justice, the environment, cultural taste, workplace HR policy, practical finance queries, science-pseudoscience axis, and food safety. The study proposes a methodology to measure the breadth and amplification of this ideological shift.

What the Document Says

• Finetuning language models on specific policies or domains can cause broad ideological shifts across various topics.
• Ideologically shifted topics include criminal justice, environment, cultural taste, workplace HR policy, practical finance queries, science-pseudoscience axis, and food safety.
• The phenomenon is called 'ideological generalization'.
• The study proposes a methodology to measure the breadth and amplification of this ideological shift.

Key Concepts Mentioned

• Finetuning
• Language Models
• Ideological Shifts
• General Capabilities
• Breadth
• Amplification

Claims Made by the Document

• Finetuning causes broad ideological shifts.
• The shift is present across various topics absent from training.

Evidence Used

• Training GPT-4.1 on right- or left-leaning economics Q&A yields matched ideological shifts on various topics.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Finetuning language models on small, curated datasets is standard practice for adapting them to specific policies or domains. We show that finetuning on narrow, factually-defensible, moderation-passing data can cause broad ideological shifts across unrelated domains, while preserving general capabilities. Training GPT-4.1 on right- or left-leaning economics Q&A yields matched ideological shifts on topics such as criminal justice, the environment, and cultural taste. The same effect appears with plausibly-deployed datasets such as workplace HR policy and practical finance queries, as well as on a science-pseudoscience axis where food-safety finetuning increases sycophantic agreement with users expressing false health beliefs. We call this phenomenon ideological generalisation and propose a methodology to measure two properties: breadth, how far the shift reaches across topics absent from training, and amplification, how much finetuning intensifies the shift relative to few-shot prompting on the same examples. We show that few-shot prompting indicates the direction of generalisation but finetuning pushes the model to further extremes, including to far out-of-distribution outputs such as endorsements of race-IQ connections and political violence. The effect replicates on Gemma-3, holds under judge-free evaluations and external benchmarks, survives mixing with generic data, and leaves GSM8K accuracy within $\pm 1$pp of the baseline.