Private Claude Chat

Claude is the smartest model out there. It's also the one most people don't realize is keeping a copy of every word.

What Claude.ai actually does with your chats

Claude is an excellent product. Anthropic is, as AI labs go, one of the more thoughtful ones about safety and policy. But "thoughtful" is not the same as "private." Here's where your data actually goes every time you type something into the consumer Claude.ai web app or desktop app:

• Your message is stored. By default, indefinitely, on Anthropic's servers, attached to your account.
• On Free, Pro, and Team plans, it can be used to train future models unless you've turned that off in your privacy settings. The default is on.
• It can be subpoenaed. If a court orders Anthropic to produce records, your chats are records.
• It's accessible to anyone who breaches your account. Compromised password, SIM swap, malware on your device. Once they're in, the entire history is in.

None of this is hidden. Anthropic spells it out in their privacy center. The problem isn't that they're doing something dishonest. The problem is that the gap between "this is technically disclosed in a settings page" and "the average user understands what they've signed up for" is enormous.

Privacy by plan

Anthropic's privacy posture is wildly different depending on which Claude product you're paying for. Worth knowing where you actually sit.

That row at the bottom matters. The developer version is a fundamentally different deal than the website. We'll come back to it.

The Incognito catch

Anthropic shipped an Incognito mode for Claude.ai in 2025. It's a real feature. It's also genuinely misunderstood.

Here's what Incognito mode does:

• Doesn't save the chat to your visible history.
• Doesn't use the chat for training, even on plans where training is on by default.
• Doesn't pull from your previous chats or memory.

Here's what Incognito mode does not do:

• It doesn't stop Anthropic from receiving your messages. They still run the model.
• It doesn't delete the chat from Anthropic's servers immediately. Anthropic retains incognito chats for 30 days by default for safety screening and abuse detection.
• It doesn't make the chat unsubpoenaable during that 30-day window.
• It doesn't encrypt the chat in transit or at rest in a way that locks Anthropic out.

There's a developer version of Claude with totally different rules

Anthropic sells Claude two ways. One is the website almost everyone uses (Claude.ai). The other is a "key" that developers and businesses get to plug Claude into their own apps. The privacy rules on the developer version are completely different.

When developers use the key, Anthropic's defaults flip:

• Operational logs auto-delete after 7 days. Anthropic keeps a short-term log for abuse detection, then it's gone.
• Conversations are never used for training. Not by default, not opt-in, not at all. It's part of the contract, not a setting you have to find.
• No saved chat history. The developer version doesn't have a "your past conversations" feature. Each message is independent. Where the history lives is up to whatever app the developer built.

That's what Private Claude is built on. We use the developer version, but wrap a normal chat interface around it. So you get the same Claude you'd use on Claude.ai, but with the developer-version privacy rules baked in.

The full list of what's flagged

Every major AI provider (Anthropic, OpenAI, Google) runs automated safety classifiers on every prompt. This is universal and unavoidable. The good news: the categories these classifiers look for are narrow, specific, and published. Here's the complete list of what Anthropic flags, drawn directly from Anthropic's published Usage Policy:

• Child sexual abuse material (CSAM) and any sexual content involving minors
• Non-consensual sexual content depicting real people
• Weapons of mass destruction: chemical, biological, radiological, nuclear
• Manufacturing instructions for explosives or conventional weapons
• Cyberweapons, malware creation, instructions for exploiting software vulnerabilities
• Attacks on critical infrastructure (power grids, water systems, financial systems, healthcare systems)
• Incitement to violence, terrorism, or genocide
• Human trafficking, exploitation, and forced labor
• Fraud, scams, and large-scale deception schemes
• Doxxing, stalking, and unauthorized surveillance of individuals
• Election manipulation and coordinated political disinformation
• Discrimination in high-stakes decisions (employment, housing, credit, healthcare access)
• Unauthorized impersonation of real people
• Generating defamatory content about identifiable individuals
• Unauthorized practice of regulated professions (medicine, law, financial advice) without proper disclaimers

Private Claude in 90 seconds

If Sarah had used Private Claude for that tax question, here's what would have been different:

1. Her connection password, not ours. She gets her Anthropic connection password from console.anthropic.com (the same way you'd grab a password for any service). It's the secret that lets Private Claude talk to Claude on her behalf. We never see it on our servers, never store it. It lives in her browser session only.
2. The conversation lives only in her browser tab. Nothing is saved to our servers. Nothing is saved to her device. Close the tab and the conversation is gone. We don't keep chat history because chat history is the thing that creates risk.
3. The message goes directly to Anthropic. Claude itself sees it because Anthropic runs the model. Anthropic's operational logs hold the request for 7 days for abuse detection, then auto-delete. They don't train on it, and there's no saved chat history at Anthropic. So a week after Sarah's tax conversation, there's no permanent record of it anywhere.
4. Nothing to subpoena. Nothing to breach. Nothing to expose. Three months later, when Sarah remembered the tax question, there was nothing anywhere for anyone to find. Not in our database, because we don't have one. Not in her account, because she doesn't have a saved history. Not at Anthropic, because the API logs auto-deleted a week after the chat.

That's the whole architecture. There's no clever cryptography trick, because we don't need one. The privacy guarantee is structural: the data doesn't exist after you close the tab.

Honest tradeoffs

Private Claude is not the right tool for everyone. Here's where it costs you something.

The free tier is for trying, not for living on. You get 50 Haiku messages and 25 Sonnet messages, using your own Anthropic connection password. That's enough for two or three real conversations. Once you hit the cap, you upgrade or stop. There's no recurring free quota.

Paid tiers are unlimited. Once you upgrade to Basic ($17) or Pro ($37), there's no PrivateClaude-imposed message cap. You can chat as much as you want, all day. The only limit is what you choose to spend on your own Anthropic account. Most casual users spend about $3 to $5 a month with Anthropic on top of the subscription.

This isn't a replacement for ChatGPT or Claude Pro. It's an addition. Most Private Claude users keep their main ChatGPT or Claude.ai subscription for everyday work and add Private Claude for the conversations they wouldn't put in their main account. The work email and the private journal aren't the same thing. ChatGPT for one. Private Claude for the other. Total cost for that setup is roughly $20 for ChatGPT or Claude Pro, $17 for Private Claude, and another $3 to $5 in API spend. About $40 a month for everything you do, plus a designated private space when you need it.

You don't get chat history. By design. When you close the tab, the conversation ends. No saved threads. No scroll-back to last week's chat. No exporting old conversations on the free tier. If you need to keep an answer, copy it before you close the tab. This is the privacy tradeoff: history is risk, and we don't store it.

You won't get features Anthropic ships first to Claude.ai. Claude.ai sometimes gets new features (like a new model, or a new tool integration) ahead of API availability. You'll usually get them within a week or two. If you need bleeding-edge first-day-of-release access, Claude.ai Pro is going to be a half-step ahead.

Anthropic still runs the model. Private Claude does not give you privacy from Anthropic's model servers. If your threat model is "Anthropic itself is the adversary," you need a self-hosted open-source model, not Private Claude. We cover the self-hosted option here.

Who this is for

You're probably a Private Claude user if any of these are true:

• You use Claude for sensitive personal stuff (health, money, relationships, work conflicts) and the idea of those chats sitting in a database for years bothers you.
• You want to try before you pay. The free tier is 50 Haiku + 25 Sonnet messages, using your own Anthropic connection password, no card.
• You want a designated private space, not a replacement for your main AI. Private Claude is $17 a month. Most people add it alongside their existing ChatGPT or Claude.ai subscription rather than swap it in.
• You want a chat interface, not a developer tool. Open API consoles and API calls are not your thing.
• You don't want to run a local model. You like Claude specifically because it's the best, and a self-hosted Llama is a real downgrade.

You're probably not a Private Claude user if:

• You only use Claude for low-stakes work and you don't care about privacy.
• You need on-device, fully air-gapped AI. Use Ollama or LM Studio with Llama or Mistral.
• You need a BAA for HIPAA-covered work. We have a Business plan for that, but the standard product isn't BAA-backed.