ITLine

ITLine AI+ / AI+ Privacy Gateway

ITLine AI+

AI+ Privacy Gateway: customer data in Claude, under code names

The AI gets a code name, the answer brings the real name back

Before the text goes to Claude, it replaces names and identifiers with code names on your company’s own machine.

Register now, and we e-mail you when we open.

Register now (in Hungarian)

Meanwhile: the free AI+ modules

The AI+ Privacy GatewayWith Hungarian narrationWatch on YouTube ↗

What is it?

AI+ Privacy Gateway runs on the company’s own machine, between Claude and the company’s texts. Before the text goes to the AI, it replaces the names and identifiers it recognises with code names, and writes them back in the answer, with correct Hungarian inflection. It is one of the ITLine AI+ modules, by monthly subscription.

The situation

For a quick answer, the whole customer e-mail goes into ChatGPT or Claude: name, TAJ number (the Hungarian social security number), bank account number.

The provider then handles the data on the company’s behalf, and the responsibility stays with the company. You choose between two bad roads: ban AI, or not know what goes out.

TODAYwith name, TAJ numbercustomer e-mailAI providerAFTERwith code namescustomer e-mailon-site gatewayAI providera manual step
The same e-mail, with one more stop. The stop is at the company.

How it works

Claude Code or Claude Cowork sends the request to the gateway, not directly to the provider.

At the company, the gateway replaces the names, addresses, company names and identifiers it recognises with code names, for example the TAJ number, the tax ID and the bank account number. The key for swapping them back stays on the company’s machine.

The AI works on the text with code names, using the company’s own Anthropic account.

In the answer, the gateway writes the real names back, with Hungarian inflection: “Nagy Péterrel” instead of “[SZEMÉLY_2]-vel”.

If the AI writes a file to your machine, the real name goes in. If a command would send data to the network, the code name stays in it.

Where the gateway is unsure, it flags the word for review, and a person decides.

What is different

  • Recognition runs at the company, with rules and local recognition models. The cloud AI gets the text with code names.
  • Built for Hungarian text: TAJ number, tax ID, company registration number, bank account number and the Hungarian inflection of names.
  • Always up to date: we teach the rules, they update continuously and we optimise for the latest models.

Common questions

What does the gateway mean for GDPR (the EU General Data Protection Regulation)?

The gateway pseudonymises, it does not anonymise. Data with code names remains personal data for the model provider too, who is the company’s data processor. The gateway reduces how much identifiable data reaches the provider. It does not replace the controller’s decisions, the notices or the contract with the provider, and this is not legal advice.

Does using the gateway exempt me from GDPR obligations?

No. Pseudonymised data remains personal data, so the contract with the provider, the privacy notices and the records are still needed. The gateway reduces how much identifiable data reaches the provider, and makes it easier to show that the measures under GDPR Articles 25 and 32 have been taken. The indirect details left in the free text (age, profession, situation) still need separate attention.

Why the gateway if we already have a Claude business plan (Team or Enterprise)?

The business plan settles the contract: Anthropic is your data processor and does not train on your data by default. It does not decide how much identifiable data reaches it; data is stored in the US, and sending only what is needed is your job. GDPR Articles 25 and 32 name pseudonymisation; that is what the gateway adds. Strong where health data or a confidential case goes through; with mostly company data it is often enough. This is not legal advice.

Can I paste customer data into an AI assistant?

If the text contains customer data, the provider handles it on the company’s behalf, and the responsibility stays with the company. Whether it is allowed depends on the company’s data protection rules; we give no legal advice on that. The gateway reduces how much identifiable data leaves the company.

We work with companies. Is our partners’ data also personal data?

The company’s name and general contact details are not. The contact person’s name, direct e-mail address and phone number are, and so are the name, tax number and address of a sole trader. With a private-person customer, every bit of customer data is personal data.

Which AI does it work with?

Today we have tried it with Claude Code and Claude Cowork, using the company’s own Anthropic account. Not yet with other clients; once we have tried one, we will say so here.

What do I need?

A machine at the company that runs Docker (a container runtime). We measure on a Mac mini with 16 GB of memory. There is no Windows installer yet.

Can it be switched off?

Yes. The administrator can temporarily switch it off in the management interface; the gateway then forwards without changes, and the interface shows this.

What reaches ITLine?

Your texts do not: the gateway works at the company, and sends the request to the provider with the company’s own account. We only receive the few technical details needed to refresh the licence (licence and instance ID, version).

When can I try it?

You can register now: register (in Hungarian). We e-mail you when download and subscription open.

What is not included

It is not legal advice, and does not replace the controller’s decisions, the notices or the contract with the provider.

There is no Windows installer yet, and we have not tried it with other AI clients.

It does not replace an outbound firewall: the AI agent can read your files on your machine.

Get rid of ExcelSee what we suggest