Remember when signing up for a new tool meant clicking a checkbox that said “I am 13 or older” and nobody, anywhere, ever checked? That checkbox was the internet’s handshake agreement with itself. Everyone knew it was theater. Everyone did it anyway.
That era is closing. As of 2026, Claude is restricted to users over 18. Anthropic drew a line, and the line is an age.
The company Claude keeps
The timing is what caught my attention. TikTok has also restricted its beauty filters to users over 18, citing children’s mental health. Two very different products, two very different companies, same conclusion: some things should have an adult on the other end of them.
I review toolkits for a living, which means I spend most of my week finding out where products break. Age gates are usually the least interesting part of a signup flow — a speed bump between me and the API key. But an age gate is also a statement about what a company thinks its product does to people. Beauty filters got gated because of what they do to how kids see themselves. Claude got gated because… well, that’s the part worth sitting with.
What else Anthropic has been up to
The age restriction doesn’t land in a vacuum. A few other things have happened around Anthropic recently, and they read like a company tightening every valve at once.
- The entire source code for Claude Code was accidentally leaked online, the result of a packaging error during a software update.
- Anthropic cut off the ability to use Claude subscriptions with OpenClaw and other third-party AI agents, closing a door a lot of people were walking through.
- Anthropic has described a model as “a step change” while warning it poses cybersecurity risks that could outpace defenders, and access to it is currently restricted.
Put those next to each other and the 18+ rule stops looking like a compliance footnote. It looks like part of a pattern: fewer doors, tighter access, more control over who touches what.
My honest read as a reviewer
I’ll say the uncomfortable part. Some of this is safety and some of this is business, and from the outside you often can’t tell which is which.
The third-party agent cutoff is the clearest example. There’s a defensible safety story there — you don’t know what an outside agent is doing with your model, you can’t audit the prompts, you can’t see the volume. There’s also a very obvious commercial story: subscriptions were being used as cheap fuel for other people’s products. Both stories are true at the same time. Anthropic gets to pick which one to lead with.
The age gate is cleaner, morally speaking. Nobody’s revenue improves by turning away teenagers. If anything, a general-purpose assistant loses a large, enthusiastic, homework-having user base by putting up that wall. That’s a cost, not a play.
But it’s also an admission. A company doesn’t gate its own product at 18 because the product is harmless. It does it because there’s some scenario it doesn’t want to defend in a hearing room, and it decided that scenario is easier to prevent than to moderate. Given a leaked codebase and a model the company itself is describing in terms of cybersecurity risk that could outrun defenders, I understand the instinct.
What this means if you’re building on it
Practical notes, because this is a toolkit review site and not a philosophy blog:
- If your product has any users under 18, Claude as an underlying assistant is now a compliance question, not a technical one. Sort that out before you sort out prompts.
- If your workflow depended on routing a Claude subscription through a third-party agent, that path is closed. Plan for official API access instead of hoping the old route quietly reopens.
- Assume more restrictions, not fewer. The direction of travel over the past several months is one-way. Build for the version of the platform that has more gates than today’s.
The part I actually like
Age gates on the internet have historically been a joke, and jokes are easy to ignore. A real one costs a company something. Anthropic and TikTok both chose to pay that cost, on completely different products, for reasons that rhyme.
I’d rather use a tool from a company that occasionally tells me no. The alternative is a tool that says yes to everything and lets the consequences arrive later, which describes most of the software I’ve reviewed in the past two years.
Just don’t confuse a locked door with a solved problem. The 18+ rule tells you who can sign up. It says nothing about what the thing on the other side is capable of, and Anthropic’s own language about step changes and outpaced defenders suggests they’re still working that out too.
🕒 Published: