Yes, Character AI monitors conversations, mostly through automated systems. The company's published safety posts describe classifiers that screen model replies, filters that block inappropriate user input, blocklists for user-created characters, and a stated practice of keeping chat data to train and improve its models. Whether any staff member reads a particular private chat is not documented in the sources that could be checked.
Key takeaways
- Automated classifiers screen what the chatbot says and block certain user inputs, according to company posts from 2024.
- The company says it keeps certain shared data to train and improve its models; an opt-out is described for users in the European Economic Area (EEA) and the United Kingdom.
- On October 29, 2025, the company announced that open-ended chat would end for users under 18 no later than November 25, 2025, paired with age assurance tools.
- The Federal Trade Commission (FTC) sent information orders to seven companies, including Character Technologies, on September 11, 2025.
- Monitoring for safety and data use for training are different things, and the company's pages treat them separately.
What "monitored" can mean
The word covers three separate activities. The first is safety filtering: software that checks messages as they are sent and received. The second is moderation of public content, such as characters that other users create and share. The third is data retention and reuse, where stored chats feed model training or product features. A service can do all three without a person ever reading a private conversation, and a person can read a chat only when a rule or report triggers a review. Each of these deserves its own question, and the company's public statements answer them unevenly.
Safety filtering: what the company has published
A post dated October 22, 2024, describes "improved detection, response, and intervention" for inputs that break its Terms of Service or Community Guidelines. It says the company uses industry-standard and custom blocklists, removes flagged characters, and shows a pop-up that points to a suicide prevention line when users type certain self-harm phrases.
A second post, dated December 12, 2024, adds detail for teens. It says classifiers screen model outputs, that new classifiers were added and existing ones strengthened for users under 18, and that separate steps block inappropriate user inputs. The company described a distinct model for teen users, a notice after an hour-long session, a reminder in every chat that characters are not real people, and extra warnings on characters with names such as "therapist" or "doctor." It also announced a parental feature that shows time spent on the platform and the characters a teen talks to most.
Data use and model training
The company's model training page says it keeps certain data users share "to help train and improve" its AI models, and that it reduces personal information in training data first. A user in the EEA or the United Kingdom can switch off the "Improve the Model for Everyone" setting, after which new content is not used to train its generative models. Even then, the page says data may still improve search, recommendations, and safety classifiers. The page describes the opt-out only for those two regions. Readers in Canada and the United States should not assume the same switch applies to them and should check the current settings and privacy policy in the app.
Where the layers stand
The table lays out what each layer covers and which date the supporting statement carries. It is a summary of company statements, not an audit.
| Layer | What is described | Source date |
|---|---|---|
| Output filtering | Classifiers screen model replies; extra classifiers for users under 18 | December 12, 2024 |
| Input blocking | Inappropriate user inputs blocked; self-harm pop-up | October 22 and December 12, 2024 |
| Character moderation | Blocklists; proactive removal and removal after user reports | October 22, 2024 |
| Parent visibility | Time spent and most-used characters for a linked teen | December 12, 2024 |
| Age assurance | In-house model plus third-party tools, including Persona | October 29, 2025 |
| Training use | Shared data kept to improve models; opt-out described for EEA and UK | Undated model training page |
The under-18 change and regulatory attention
On October 29, 2025, the company said it would remove open-ended chat with its characters for users under 18, with the change taking effect no later than November 25, 2025. Teens would still be able to create videos, stories, and streams with characters, and a two-hour daily chat limit applied during the transition. The same post said the company would fund an independent non-profit, the AI Safety Lab. The announcement did not describe a new monitoring system.
Separately, the FTC announced on September 11, 2025, that it had issued orders to seven companies, among them Character Technologies, Alphabet, Meta Platforms, OpenAI, Snap, and xAI, asking how they test chatbots acting as companions, how they limit harm to children, and how they monitor compliance with their own rules, including age limits. The orders are a request for information. They are not a finding against any company.
Practical steps for users and parents
- Treat chats as stored data, and avoid sharing full names, addresses, school names, or health details with any chatbot.
- Read the current privacy policy and the in-app data settings, because policies and settings change.
- Remember that a filter reduces risk but does not guarantee that harmful replies never appear.
- Parents of teens can discuss what the characters are, and what the notices in the app mean.
This article summarizes public company statements and one regulator announcement. It is general information, not legal, privacy, or mental health advice. A person in distress should contact a local crisis service or a health professional.
FAQ
Can Character AI staff read private chats?
The checked sources do not say. They describe automated screening and review triggered by rules, and the company's help pages could not be opened during verification. The privacy policy in the app is the place to confirm current practice.
Are conversations used to train the AI?
The company says it keeps certain shared data to train and improve its models. Users in the EEA and the United Kingdom have a described opt-out that stops new content from training its generative models.
Can teens still chat with characters?
The company announced on October 29, 2025, that open-ended chat would end for users under 18 by November 25, 2025, and that teens would keep creative features. Current behavior should be checked in the app.
Is Character AI under government investigation?
The FTC issued an information-gathering order to the company in September 2025 as part of a study of companion chatbots. That is an inquiry, not an enforcement action.
Related reading
- What Character AI is and how it works
- The Character AI beta app
- Character AI and Google Cloud
- How Canada regulates AI
More in AI Blog.
Last verified: October 2026
Sources
- Taking Action to Protect Under-18 Users, Character AI blog, October 29, 2025
- FTC launches inquiry into AI chatbots acting as companions, Federal Trade Commission, September 11, 2025
- How Character AI Prioritizes Teen Safety, Character AI blog, December 12, 2024
- Community Safety Updates, Character AI blog, October 22, 2024
- Model training page, Character AI, 2026