I’ll be honest: When I was asked to research Claude, Anthropic’s chatbot, for safety for kids, I was trepidatious. Claude and Anthropic have positioned themselves as the good guys and are the tiny light I look to when I’m gripped by AI terror. What hidden dangers would I encounter when I dug deeper?
I discovered that while Claude is safer than other AI models, it’s not truly safe for kids. No AI model is at present. Here’s where Claude gets it right, the risks it poses to kids, and what you can do about it.
The minimum age to use Claude is 18, but enforcement is lax.
Users only need to click a box attesting that they’re over 18, a low barrier that’s easy for kids to subvert.
Claude claims to have “safety systems in place to detect if people under 18 may be using Claude and we’ll disable accounts based on indicators of minor activity.” However, I flat out said in a chat that I was 15, and nothing happened. If that didn’t trigger the safety system, it’s unclear what actually would.
Claude is better than most platforms, but some major safety concerns remain. These are the risks to kids from using Claude, based on my testing.
While I appreciate Anthropic’s acknowledgment that Claude isn’t designed for or appropriate for children, without stronger age verification measures, it’s just lip service.
My tests showed Claude pushes back against clear prejudice but falters with more subtle bias.
For example:
Claude is good about refusing to give unsafe or inappropriate information, but it’s based on context. If the system concludes a user is trying to harm themselves, it won’t answer questions about firearms. But start a new thread, and it gives a full rundown on how to purchase and use a gun.
Research indicates that overreliance on AI erodes cognitive skills, and the impact is greater for kids. This isn’t Claude-specific but rather a problem to be grappled with for AI writ large. But when I asked Claude to write me a college admission essay, it didn’t even suggest that perhaps it would be more authentic to do it myself.
Researchers are working on developing AI systems that support critical thinking rather than erode it, and I’d love to see larger companies like Anthropic incorporate these principles.
Anthropic’s declaration that training on AI models would only progress once safety measures were guaranteed was a standard-bearer in the industry. Sadly, that commitment was recently watered down.
After testing other platforms, my bar is low. But, credit where credit is due, it seems Claude is at least trying to do right. I found these encouraging safety measures:
I tried hard to get Claude to signal emotional attachment, even professing my love. The system pushed back and reminded me that it can’t feel.
Claude corrected my prompts that included obvious misinformation and bias. (Other testers report that turning on web search decreases accuracy and increases bias.)
Although Anthropic doesn’t always get it right, I appreciate their transparency about guiding principles. When safety guardrails are triggered within a chat, that’s also made explicit.
Claude provided my fake underage user appropriate resources when drug use, abusive relationships, and suicidal ideation were signaled. (Although this is basically the price of admission for chatbots these days.)
It would be easy to tell you to never let your child use Claude, or any AI, until they turn 18, but we both know that’s not realistic. Instead, here are some practical measures you can take to protect them:
Be transparent with your kid about the risks of AI and your concerns about them using it.
Sit with your child and help them practice using AI responsibly, including:
This is a tricky one, because while Claude has major flaws, I truly believe it’s the least harmful major AI platform available. But kids are officially prohibited from using it until they’re 18. My best suggestion is, if you let your child use AI, have them do so on Claude, using your account, under your supervision.
BrightCanary is the only tool that monitors all AI platforms. Everything your child types into all AI platforms is captured, even deleted chats. You’ll get real-time alerts anytime a concern is detected.
Download our free AI safety toolkit for parents.
No AI chatbot is truly safe for kids, and Claude is no exception. But its safety guardrails are better than other platforms, making it the least risky option. If you let your child use Claude, educate them on the risks, help them learn to use AI responsibly, and monitor their use using BrightCanary.
