Roblox Shares New AI Tools to Keep Kids Safe Online

Roblox co‑founder and chief executive Dave Baszucki talked about new safety tools on the TV show Fox & Friends. He explained the new Kids and Select account types, said the company is fighting lawsuits and worries about online predators, and described age checks, content filters and tight chat rules to keep users safe. Baszucki added that Roblox will not allow bad behavior and that parents can change chat settings for their kids.

If your child plays Roblox, many of their friends probably play it too. Roblox said it had about 123 million daily active users in the second quarter of 2026. Almost three‑quarters of the users who have their age checked are under 18. With so many young players, a big danger can start with a simple chat that looks harmless. One person may ask what games the child likes. Later they may ask where else the child chats. Then they may suggest moving the talk to another app. Parents who have seen online grooming cases will recognize this pattern.

Earlier this year we reported a scary Roblox safety case involving two girls, ages 12 and 14. Investigators said a man from Nebraska first talked to them on Roblox, then moved the chat to Snapchat. Police called the case online grooming. Roblox now says it will share some of the AI it uses to find risky behavior.

A former teacher and GOP lawmaker is accused of using Roblox to lure a student into a sexual relationship.

Roblox is releasing new AI safety tools that can spot grooming signs, requests for personal data, and unsafe voice chats involving children.

On August 19 Roblox said it will add three updated safety models to the ROOST (Robust Open Online Safety Tools) Model Community. It will also share a new test dataset so other platforms can try their own safety systems. For parents the technical details matter less than what the tools try to catch. They look for attempts to get personal information, signs of possible child danger, and bad voice‑chat behavior. This article explains how the new AI tools work, what they could mean for your child, and what you can do now to keep kids safer online.

New free live CyberGuy class teaches how to protect your money from major threats.

Reserve your free spot today. Visit CyberGuyLive.com.

Roblox already uses these AI systems on its own site. Now other companies can study them and maybe use them for their services. Roblox joined ROOST as a founding member in 2025 with Google, OpenAI, Discord and others. ROOST aims to give open‑source safety tech to groups that cannot build complex systems themselves.

Think about how many places children chat online today. A talk can start inside a game. Then it may jump to a messaging app or another social site with different rules. No single company controls the whole journey.

Sharing safety tech could let more services use tools that look for the same warning signs. But each company still has to add the tech to its own platform. So this is a good step, but parents should still stay alert and not lower their guard.

One update for families is Roblox’s PII Classifier. PII means personally identifiable information, such as a phone number, email address, social‑media username or other data that could help someone find or contact a child.

The classifier also checks for attempts to move players to other platforms. That matters because a bad person might try to take a conversation away from the safe place where it began.

Older filters often looked for obvious words or patterns. Roblox’s Version 2.0 looks at the whole conversation instead. For example, someone might misspell the name of another app, split contact details across several messages, or use coded language. The new AI evaluates those messages together.

Roblox says Version 2.0 now supports 189 languages, up from 17. Its F1 score – a common accuracy measure – rose from 63.41 to 90.52. For parents, the simple point is that the system tries to see where a chat is heading, not just judge each message alone.

One part of the news that stood out to me is that asking for another username or suggesting a different app can sound innocent. Kids switch between platforms all the time. Still, moving a chat can reduce the protections that were in place.

That is why the Nebraska case we mentioned earlier felt so disturbing. Authorities said the first contact happened on Roblox before the conversation moved to Snapchat.

Parents should talk about this at home. If someone the child only knows online asks for another username, phone number, or a private way to talk, the child should tell a parent.

Roblox Sentinel deals with a harder problem. Grooming can grow slowly. Someone may earn trust before the chat becomes clearly bad. Early messages can look normal when looked at alone.

Sentinel looks at patterns across chats. Roblox says the system searches for early signs of possible child danger so risky talks can be reviewed before they get worse. Human reviewers can then focus on conversations that may need action.

Roblox says that in the 12 months ending August 7 2026, almost 70 % of the child‑danger cases it found came through Sentinel’s early detection.

Parents push Congress to act on kids’ online safety after juries find Meta and YouTube liable for harm.

Roblox Sentinel looks at chats for early signs of possible child danger, while human reviewers can prioritize interactions that may need action. (Roblox)

Roblox is also releasing Sentinel Version 2. The changes mainly help developers and safety teams who will use the tool. The new version gives them more ways to score suspicious behavior and decide which setup works best for their platform.

Roblox says Version 2 can also show more detail about why something got a certain score. That can help teams understand what the system saw instead of only getting a vague warning.

The company says the upgrade also makes the system much faster to test and tune. For parents, the big point is that Roblox wants other platforms to take this technology and use it for their own safety systems.

Text messages are only one part of the picture. Kids can also talk using voice chat, which creates a different moderation challenge. Roblox’s voice safety classifier checks speech for policy breaks in real time. When it finds a violation, Roblox can show a warning explaining which rule may have been broken.

Repeated violations can lead to a temporary voice‑chat suspension of up to five minutes. More serious breaks can bring stronger penalties. Roblox says its voice safety classifier has been downloaded more than 72,000 times since it was open‑sourced in 2024. Version 3 now covers 30 languages and eight violation types. Roblox reports a 61 % recall across those languages at a strict 1 % false‑positive rate. This reminds us that AI moderation can miss things; even a sophisticated model cannot catch every dangerous conversation.

There is another side of this story worth knowing. Independent researchers recently examined Roblox chat moderation and found unsafe messages that slipped past the platform’s existing safeguards. Their study looked at more than 2 million chat messages and found examples of grooming, sexualizing minors, bullying, violence and sharing sensitive information that escaped moderation.

That research shows why parents should not assume automated moderation is a complete safety net. Bad actors learn how filters work, then change their language or behavior to get around them. This makes Roblox’s latest move especially important, because the company is improving these safety models and sharing them with other platforms so the technology can keep adapting too.

This may become the bigger story. Most parents will never download an AI classifier or visit a GitHub repository. Yet the technology released today could eventually affect apps their children use. Smaller platforms may not have the engineering teams or money needed to build advanced child‑safety systems from scratch.

Open‑source tools can give those companies a starting point. Researchers can also test the models and look for weaknesses. Developers can adapt them to different services. Meanwhile, feedback from other groups could help improve the original tools. ROOST says its Model Community is meant to connect developers and safety experts around open safety AI models. That collaborative approach has a lot of potential.

This isn’t Roblox’s first big AI safety push. Earlier this year, CyberGuy reported on a Roblox system that looks at text, avatars and 3D worlds together in real time. Roblox said the technology helped shut down about 5,000 violating game servers each day. The idea was to catch harmful combos that older moderation systems could miss when they looked at each piece separately. You can read our full report on how Roblox is changing online safety with AI.

Roblox has also been expanding age‑based protections for younger users. Those controls set different limits on chat, games and other features depending on a user’s age. All of these moves aim at the same goal: catching risk earlier and limiting who younger users can interact with. The challenge is making those systems work consistently at Roblox’s huge scale.

Chris Hansen urges parents to teach children about online predators before allowing internet access.

Roblox’s latest child‑safety AI looks beyond single messages to find patterns that could signal attempts to move children to other platforms. (JasonDoiy / Getty Images)

These AI tools work behind the scenes. Parents can add another layer of protection by setting clear expectations at home.

Tell your child to come to you when someone they only know online wants to keep chatting somewhere else. That could be another gaming service, a social platform or a private messaging app. There may be an innocent reason, but it is worth knowing when someone tries to move the conversation.

Children should be careful about sharing information that can identify or locate them. That includes their phone number, school name and home address. Social‑media usernames can also give strangers another way to contact them. Remind your child that a person they meet inside a game is still an online stranger until your family knows otherwise.

Roblox has expanded its age‑based account protections and parental controls. Roblox Kids accounts cover ages 5‑8, while Roblox Select covers ages 9‑15. Chat options and parental controls differ by age group. We previously broke down Roblox’s age‑based accounts for kids and teens and what parents should know about those changes. Take a few minutes to review the settings with your child rather than assuming the defaults match what you want.

Parents naturally focus on text because it leaves something visible to review. Voice conversations deserve attention too. Ask your child who they talk with while gaming. Make sure they understand that they can leave any conversation that feels uncomfortable.

Roblox lets users report people and bad behavior from within a game. Users can also block another account. Parents can reach communication controls through linked parental settings. Show your child where those controls are before a problem occurs. That way they do not have to figure it out while they are upset or scared.

Your reaction matters when your child tells you something went wrong online. Try to make the first talk about what happened rather than immediately focusing on taking the game away. A child who thinks speaking up will automatically cost them their phone or Roblox access may be more reluctant to tell you about the next uncomfortable interaction.

What catches my attention here is what these AI systems are actually being trained to recognize. They are looking for someone trying to get a child’s personal information or shift a conversation somewhere else. Another system examines chats for early signs of possible child endangerment. Voice moderation adds protection when kids are talking instead of typing. Those are situations parents need to understand. I also like the idea of companies sharing safety technology instead of making every platform build these tools from scratch. If technology developed on Roblox helps another gaming service or social platform detect dangerous behavior sooner, that is a positive step. Still, I would never tell a parent to assume AI has their child’s back. Roblox itself acknowledges that no safety system is perfect, and independent researchers have found examples of harmful chat getting through moderation. So use the technology as another layer. Then keep talking with your kids about who they meet online and where those conversations are going. The most important warning may be something an algorithm never sees: your child telling you that somebody online is making them uncomfortable.

If an AI system detected possible grooming in your child’s online conversation, how quickly would you expect the platform to alert you, and how much should it tell you about what happened? Let us know by writing to us at Cyberguy.com.

CLICK HERE TO DOWNLOAD THE FOX NEWS APP

Kurt "CyberGuy" Knutsson is an award‑winning tech journalist who loves technology, gear and gadgets that make life better. He writes for Fox News & FOX Business and appears each morning on "FOX & Friends." Got a tech question? Get Kurt’s free CyberGuy Newsletter, share your voice, a story idea or comment at CyberGuy.com.

COPYRIGHT DISCLAIMERS

4English does not own the rights to this content. All rights belong to the owner. 4English only uses this content to help users learn English via the app's features. Contact to request the removal of your content: contact@4englishapp.com.

Luyện nghe