content moderation

basicwidespread

Content moderation is the checking and filtering of whether text, images, or videos appearing on an online platform or in an AI system comply with rules and laws.

In plain terms

Imagine a nightclub doorman and a security guard walking the floor. One decides who gets in; the other intervenes if someone causes trouble inside. Content moderation does the same on the internet: it monitors what is posted on a platform and removes or restricts anything that violates policy or law.

Originally, the task of social media platforms, forums, and comment sections was to filter out hate speech, harassment, fraud, violent content, or material endangering children. Today, machines largely perform the initial screening, while human moderators handle the difficult, disputed cases.

In the AI era, the concept has expanded. For chatbots and AI agents, you must filter both what users request from them and what they respond or do themselves. Thus, the machine's own output also requires moderation, alongside that of humans.

In moderation, the platform enforces its own published rules, which can be stricter than the law. Censorship in the classical sense is a state ban. However, the line can blur if the platform removes content under state pressure or regulatory instruction.

How it works

Rules: the provider describes what is forbidden and what is allowed (community guidelines, terms of use) and adjusts the filtering accordingly.
Machine pre-screening: classifier models score content by category (such as violence, harassment, fraud), and known prohibited images are immediately recognized based on digital fingerprints.
Decision: anything clearly violating rules is deleted, hidden, access-restricted, or marked with a warning label; in an AI assistant, this corresponds to refusing the request or withholding the response.
Human review: uncertain cases and user reports are examined by moderators who know the language and cultural context.
Appeal and learning: the affected party can object, and the machine filter is further refined based on corrected decisions.

An everyday example

You upload a video, and within minutes you receive a notification that it has been age-restricted or removed because the system deemed it a violation. Something similar happens when an AI chatbot politely rejects your request: this is triggered by a separate filter or rules trained into the model.

Who is affected

everyoneWhat you see or do not see on social media, and what an AI assistant is willing to answer, is largely shaped by moderation decisions.

parent, student, teacherFiltering harmful content for children largely depends on this, but since the filter is not perfect, it does not replace supervision. → Education & Family

entrepreneur, executiveIf your company has a comment section or rating system, you are typically liable for infringing content uploaded by others only if you become aware of it and fail to act. However, responses from your AI agent communicating with customers count as your company's own statements, for which you may be liable without prior notice.

developerWhen building an AI application, you must implement filters to check inputs and outputs, using either your own or a ready-made moderation service, and you must decide the thresholds. → Code

A common misconception

MythModeration is now handled entirely by artificial intelligence, which accurately identifies rule violations.

In realityThe machine filters mass, clear-cut cases well, but often misunderstands irony, slang, context, and smaller languages. Human intervention is still needed for contentious decisions, and errors occur regularly in both directions.

Limits and risks

It can make two types of errors: blocking harmless content while letting actual violations slip through, requiring a constant compromise between the two.
On smaller languages, including Hungarian, machine filters typically perform worse than in English.
Rules are set by the platform, so the same content may be allowed in one place and prohibited in another, and the reasoning behind decisions is not always transparent.
AI system filters can be bypassed with clever phrasing, so moderation reduces risk but does not eliminate it.

Similar, but not the same

censorship – Censorship in the classic sense is state prohibition or prior review. Moderation is a provider's enforcement of its own public rules on its own platform, though the boundary between the two is disputed under state pressure.
guardrail – Guardrails are broader: they encompass all behavioral and action limits of an AI system, of which content filtering is just one element.
fact-checking – Fact-checking examines whether a statement is true, while moderation checks whether content is allowed on the platform.

Related concepts

Broader

AI safety, online safety, platform regulation

Narrower

content filter, hate speech detection, output filtering, notice-and-takedown procedure

On the edge

guardrail, jailbreak, prompt injection, deepfake, human oversight

A brief history

Moderation dates back to the first online communities (BBSs, Usenet, forums). Its legal framework in the United States was established by Section 230 of the Communications Decency Act, adopted in 1996. In the European Union, the Digital Services Act (DSA), adopted in 2022, imposes detailed obligations on platforms. It applies to the largest platforms from August 2023 and to all affected providers from February 17, 2024.

Where things stand

By the end of 2026 it is a core element of every major platform and AI service, and machine pre-screening is increasingly done by large language models. The debate now is about how strict and transparent it should be, and how it can be extended to autonomous agents.

Status: 2026 November

In the news

Appears in 3 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Meta introduces AI tools to screen ads leading to child sexual abuse materialOct 7, 2026 · Marketing
EU earmarks billions for AI gigafactories and tightens oversight of ChatGPTAug 31, 2026 · Agent Age
OpenAI launches ChatGPT for Teens with study features, parental controls, for ages 13–17Aug 24, 2026 · Education & Family

Also known as

content moderation · content moderation · moderation · moderation · content filtering · content moderation · content filtering

Sources

Wikipedia: Content moderation

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.