AIROS Gateway · Data protection for everyday AI use
Every AI request checked.
Before it leaves the company.
The AIROS Gateway is the control point between your employees and external AI services such as ChatGPT or Claude. It detects sensitive data, replaces it with pseudonyms, lets the request through and translates the reply back. With an audit trail for every single request.
How a request flows
The original stays with you.
Only pseudonyms go out.
The gateway runs on the device or in front of your AIROS applications. No cloud service, no copy of your data with us.
Allow AI in the company.
Without data leaks.
The gateway is not a wall that shuts everything down. Everything gets through, but nothing unchecked. You see which AI services actually run in-house, sensitive data stays on the device, and every request has an auditable record.
Four steps
Detect, check, pseudonymise, log.
All of it happens within milliseconds, before the request leaves the device.
Detect
The gateway recognises AI traffic by the behaviour of the request, not only by the provider's name. This makes shadow AI visible that is on no list.
Check
The data guard finds credentials, IBANs, card numbers, social security numbers, VAT IDs, health and salary data as well as names, companies and addresses. Checksums prevent an invoice number from being mistaken for an IBAN.
Pseudonymise
Instead of black bars, each finding gets a pseudonym: [PERSON_1], [COMPANY_1], [IBAN_1]. The same person keeps the same pseudonym, the AI replies sensibly, and the reply is translated back on the device.
Log
Every request is in the audit trail: type, location and number of findings, chained by checksums, tamper-evident. Never the text itself. Auditable, for example for EU AI Act questions.
Security level per provider
You decide how strict.
For each AI service you pick the level in the overview. The change applies immediately to all workplaces.
Strict
All data classes are pseudonymised, including names and companies. The right choice for law firms, HR and finance.
Standard
Credentials, accounts, ID data and health data are always protected, contextual data according to your rules.
Relaxed
Minimum protection only: credentials, cards and accounts stay off limits, names may pass because a task needs them. Visibly flagged in the overview.
For AI agents too
Hidden instructions detected.
Data exfiltration stopped.
As soon as AI agents read documents, open e-mails or operate tools, data protection for prompts is no longer enough. So the gateway also checks what the agent does.
- 01Injection guardDetects hidden instructions in retrieved documents, e-mails and web pages, including invisible control characters. A sentence in a foreign document weighs three times as much as in a user message.
- 02Agent guardClassifies tool calls by effect and recognises the classic exfiltration pattern: read a lot, then send once to a foreign address.
- 03Stops the agent, not the humanA critical run blocks the agent's automatic continuation. If the human writes on afterwards, they have decided. The incident stays visible in the overview.
Who is this for?
Wherever employees work with customer data.
The gateway is a building block of AIROS and can also be used without further AI projects.
Law firms & consulting
Client data stays in-house even when a brief is prepared with ChatGPT. Confidentiality remains enforceable.
Trade & industry
Customer names, prices and terms do not leave the company when quotes or texts are created with AI.
HR & finance
Salaries, IBANs and social security numbers are detected and replaced before they reach a cloud.
Teams with AI agents
Agents that read e-mails and operate tools get a guard against hidden instructions and data exfiltration.
Your next step
Which AI services are already in use at your company?
In one session we look at where AI is used today and set up the gateway together. In person anywhere in Austria or online.