I am an independent developer, currently unemployed. I have been using Claude Pro for development.

A recent subscription upgrade made me reconsider my relationship with the company behind it.

That morning, I tried to upgrade from Claude Pro to Max. After the upgrade, the system asked me to verify my identity. I was in San Jose, California. I did not yet have a US driver’s license, so I tried using my mainland Chinese passport.

I had already exhausted my Pro usage allowance. I could not send more messages to Claude or do any work with Claude Code. I followed the instructions and entered the verification process.

I tried about three or four times without success. The page said the photos were not clear enough, possibly because of glare or other issues with how they were taken. Up to that point, I could understand it: document verification needs a clear image, and a poor photo means trying again.

Then I received an email from Anthropic saying my account had been suspended.

Before that email arrived, I had not had any new conversations with Claude that day or done anything with Claude Code. I had just upgraded my subscription, then lost access to my account while trying to complete the requested identity verification.

I do not know whether my Chinese passport was relevant, or whether repeated verification failures triggered something. I can describe the sequence of events, but I cannot explain the relationship between them.

Anthropic’s official documentation confirms that Persona helps handle identity verification. It even includes a question about why an account might be banned after verification. But the answer gives general policy explanations, not an explanation of what happened to my particular account. Official identity verification documentation.

That is what unsettles me.

A usage policy can be long and detailed. But when a user receives a suspension decision without knowing which action corresponds to which rule, it is difficult to judge whether the decision makes sense.

If there is a real problem, I need to know what it is. If the system made a mistake, I need to know what information could establish that. Otherwise, I can only guess: was it the document, the photo, the subscription upgrade, or something I cannot see at all?

The company writes the rules, interprets them, enforces them, and controls whether the user can continue using the service. The user may see only the final outcome.

From the user’s side, this is a deeply unequal relationship. The company can act immediately when it identifies a problem. A user trying to understand what happened may not get a specific answer. That way of handling things strikes me as arrogant.

And the more useful Claude becomes, the harder it is to treat this as an ordinary customer support problem.

In my development work, Claude is genuinely difficult to replace. I use Opus 5 frequently, and its way of understanding tasks and moving work forward is often efficient.

I have called this experience a “token conversion rate”: how many tokens, exchanges, and course corrections does it take to produce a result I can actually use?

On my own tasks, Claude often gets things done with fewer exchanges. Even if Codex offers a larger allowance, that does not necessarily mean I can finish the same amount of work with the same effort. This is my personal experience, not a controlled comparison, but it does affect which tools I choose.

Switching providers therefore carries a real cost for me. I have to learn a different set of limitations, adjust how I work, and potentially accept a drop in efficiency.

The more a tool helps me do my work, the easier it is to depend on it. Yet that dependence does not automatically bring a more dependable service relationship.

That is why one account suspension led me to a larger concern.

I see July of this year as an important moment. Access to Fable 5 was restored that month, and I believe this generation of models has changed the question of how much AI can do. Fable 5 redeployment announcement.

My impression is that many of the intellectual capabilities needed for work are beginning to appear in these models. Turning that impression into a claim about how many jobs they could replace would require more evidence. But I already find it difficult to see them merely as tools for answering questions.

I think about two turning points.

The first is a capability turning point: models can handle a wide range of complex tasks and keep work moving forward.

The second is a cost turning point: that level of capability becomes affordable enough to use much more widely and frequently.

I think we have crossed, or are very close to, the first turning point. I do not know when the second will arrive. I do not have a strict definition of AGI either. But I worry that society will start experiencing the consequences before everyone agrees on what the term means.

If more and more work can be completed with these models, then who gets access, under what conditions, and when that access can be revoked become increasingly important questions.

That makes me concerned about a small number of companies controlling advanced AI capabilities.

Once a model has accumulated enough context, it can often anticipate what I want to do next. That is helpful in development. But the experience also makes me wonder what happens if the ability to understand a person is used to continually assess and restrict them.

I have no evidence that this is how my account was handled. My concern is this: when understanding users, writing rules, and enforcing restrictions all sit within the same service, what can users do to hold it accountable?

It would be inaccurate to say that Anthropic faces no oversight at all. The company has publicly explained that access to Fable was once suspended because of a government directive. But a government’s ability to restrict a model does not mean an ordinary user can obtain a clear explanation and an effective review.

One detail stayed with me. When Anthropic faced government restrictions, it publicly called for a process that was transparent, fair, clear, and grounded in technical facts. Anthropic’s statement at the time.

I understand that demand. I hope the company understands that users need those things when facing its decisions, too.

I do not know whether Anthropic intends to create a highly centralized relationship of control, and I cannot judge the motives of an entire company from one experience. But the capabilities a company controls and the ability of its users to challenge its decisions are questions that need to be considered together.

I still think highly of Claude’s capabilities. That is precisely why I care more about how Anthropic treats the people who depend on it.

If I broke a rule, tell me specifically what the problem was. If identity verification failed, tell me how that relates to the account suspension. If a decision may be wrong, give me a way to actually correct it.

It is hard to feel comfortable building more and more of my work around a decision I cannot get explained.