Concern 549 · Working draft · editorial review in progress
AI censorship at scale
AI-enabled censorship suppresses access to information or expression through automated selection, blocking or response shaping.
Understanding the concern
A system can filter large amounts of content or repeatedly omit particular viewpoints. Sometimes the restriction is visible as a refusal; sometimes it appears as an ordinary answer that leaves out contested facts. The concern is especially serious when people cannot tell that a political rule is shaping what they see.
This differs from proportionate moderation aimed at preventing abuse. The questions are who sets the restriction, what rights it affects and whether people can challenge errors. Research on tested language models has examined censorship patterns and the influence of restricted training material. Those findings should not be generalized to every model or every safety refusal.
An example
A user asks about a public event and receives a consistently incomplete account without an explanation of the restriction.
Illustrative unless identified as a documented case.
Words in plain English
- AI
- Software that learns patterns from examples or uses rules to perform tasks such as recognizing images or generating text. It can make mistakes and does not understand every situation as a person would.
- Model
- The learned part of an AI system. It uses patterns found during training to produce an answer or prediction.
- Training
- The process of using examples to shape how an AI model behaves. Learning patterns from material does not establish that the material is true or fair.
Evidence and its limits
The cited research concerns tested censorship patterns. It does not establish that every refusal is censorship or that every model behaves alike.
Amed and Knockel · PETS 2025, via Citizen Lab · Research summary
An Analysis of Chinese Censorship Bias in LLMs ↗ (opens in a new tab)Examines censorship patterns and training-data effects in tested models; does not characterize every model or user.
Broader frameworks and background
These resources provide context. They do not establish every claim in this concern.
- MIT AI Incident Tracker (opens in a new tab)
incidents
- MITRE ATLAS (opens in a new tab)
framework
- International humanitarian law (IHL)
framework
The body of law that governs the conduct of armed conflict and the protection of people who are not fighting, which applies to AI used in or for conflict.
A question to discuss
What should people be told when a political rule limits the information an AI provides?
Keep track of different views, what needs checking, and any next step in your own discussion notes.
Open discussion notes →Principles for the conversation
Before you finish · reflect together
Virtues and principles to consider
Virtues are qualities we practice in how we treat people and make choices. They are not scores for judging someone. Use these ideas to explore what a thoughtful response could look like.
Courage
Courage means acting on a well-grounded responsibility even when it is uncomfortable. It can mean trying something useful, admitting a mistake, or challenging pressure to move too fast.
Make it possible for people to raise concerns without punishment. Support those who point out harm, and be willing to pause a popular project when evidence gives a good reason.
Ask together: What important action are we avoiding because it is difficult?
Stewardship
Stewardship means caring for resources and systems that others depend on, including people who will use them later. It looks beyond the benefit of a quick launch.
Plan for maintenance, energy use, data protection, and eventual shutdown. Ask who will look after the tool when the original team moves on and how future users can avoid inheriting hidden costs.
Ask together: What are we responsible for caring for over time?
Prudence
Prudence means using careful judgment before and during action. It balances worthwhile opportunities with the need to avoid harm that could have been anticipated.
Start with a limited trial, set stopping points, and prefer choices that can be undone when uncertainty is high. Review what happens rather than assuming the original plan remains sound.
Ask together: What is a careful next step, and when should we stop or change course?
Editorial ratings
These are editorial judgments from the register, not measured probabilities or predictions about your situation.
About the register and scoring →More ways to explore
Search for more writing
For more specific or recent material, open a search below or ask for a web summary. AI-generated summaries can contain mistakes; open the original articles to check them.