We preventunintended data losswhen employees use AI.
We protect your confidential data without blocking the use of AI. When a prompt contains confidential data, that data is processed inside your security perimeter, and the answer comes from a model of your choice.
An intelligent control layer that keeps token costs in check.
BRANE decides for every prompt where it can be answered safely. An optional cloud budget per person caps token spend; once it is used up, prompts are answered locally.
Sensitive data stays inside your protected environment.
Detection, masking and logging run on your own infrastructure: on-premises, hybrid or fully offline, built around your requirements for sovereignty and governance.
Audit & Compliance
Evidence, not assertions.
Full audit logs and a tamper-evident evidence history show who sent which class of data to which model, and when. Usage analytics show how and where AI is used across your organisation.
Technology and origin
Engineered in Germany, headquartered in Stuttgart.
NVIDIA hardwareSoftware & ServicesMade in Germany
Trusted by
From mid-market to global leaders.
One control layer betweenyour people and every AI.
- Data protection
No sensitive plaintext leaves the protected environment
Personal data, contracts and trade secrets are detected and masked before the request goes out.
- Audit logging
Every interaction accountable
Who, when, which model, which data class? Logged for audit, still readable months later.
- Compliance
EU AI Act and GDPR on the record
Policies per role and data class, with the evidence that they took effect. AI-Z GmbH: ISO/IEC 27001 in progress.
- Cost control
Optionally, set a cloud budget per person. Once it is used up, prompts are answered locally.
Six functions between your people and cloud AI.
Detect, understand, mask, route, release, block. Every prompt runs through the same control layer before it leaves the protected environment: pattern detection in milliseconds, the semantic check in seconds.
Detect what nobody else sees.
Classification of every data class before sending: pattern detection in milliseconds, the semantic check in seconds. PII, IP, contracts, code, trade secrets.
Your secure AI assistant.
How can I help?
Example prompt
My name is Laura Schmidt, I live in Munich and my phone number is 0171-3920042. Please summarise my contract VTR-2024-8891.
Our browser extension spots sensitive data before the AI does.
BRANE for Browser marks personal data in ChatGPT, Claude and Gemini as you type and replaces it with placeholders on request.
Good evening
Claude can make mistakes. Please double-check responses.
Recreated interfaces. ChatGPT, Claude and Gemini are trademarks of their respective owners.
Go one level deeper.
Keynote
Watch our vision for secure AI
Your AI quick check
Three answers is all it takes. See how many prompts leave your network each year, and what that means with the provider you use today.
Risk
Your people already use AI. What risks does this expose you to?
Without central control, you cannot see what goes into AI tools. That creates four risks.
Confidential data leaves the company.
Customer data, contracts and source code can end up in prompts. Once a prompt is sent, you cannot call it back.
Shadow AI goes unseen.
Where no approved tool exists, employees often turn to private accounts. In IBM’s 2026 study of 602 organisations that suffered a data breach, the share reporting security incidents involving shadow AI more than doubled, from 20% to 43%. Breaches involving shadow AI cost USD 5.39 million on average.
Accountability stays with you.
Personal data in prompts remains your responsibility. Under the GDPR, you must be able to demonstrate compliance with the data protection principles.
Trade secrets can lose protection.
Information is only protected as a trade secret if you take reasonable steps to keep it secret. Uncontrolled AI use makes those steps hard to prove.
An EU storage location does not rule out US access.
Providers of cloud and communication services under US jurisdiction can be compelled to disclose data in their possession, custody or control, even if it is stored outside the United States. This can also apply to enterprise AI services from providers under US jurisdiction.
Transfers to the US remain legally contested.
Personal data may only be transferred outside the EEA under the conditions of Chapter V GDPR. The Court of Justice of the EU invalidated the previous EU–US framework in 2020. Its successor, the EU–US Data Privacy Framework of 2023, is in force. The General Court upheld it in 2025; an appeal is pending before the Court of Justice (as of October 2026).
Professional secrecy also binds service providers.
In Germany, lawyers, doctors, tax advisers and other professionals bound by secrecy may disclose secrets entrusted to them to external service providers only as far as necessary, and must ensure those providers are bound to confidentiality.
What changes with BRANE.
BRANE works as an AI gateway with data loss prevention for generative AI: it checks every prompt sent through BRANE before it leaves your organisation. When it detects confidential data, a model inside your protected environment answers, or only placeholders go out. Every decision is logged.
This overview is not legal advice.
News & Perspectives
Agentic Security·The Hacker News“Phishing 3.0”: Security Researchers Say the Next Wave of Attacks Is Agent Versus Agent
New analysis describes a shift from static phishing emails to AI agents that generate personalized, conversational, multi-channel lures faster than a person — or a traditional filter — can keep up with. A 2026 Dark Reading readership poll found 48% of security professionals now rank agentic AI as the top attack vector of the year, ahead of deepfakes, while independent testing found Microsoft 365 and Google Workspace's built-in email defenses are missing hundreds of phishing messages per 100 mailboxes every month.
Source: The Hacker NewsRead article
Supply chain·heise.deLiteLLM attack: over 2500 companies affected
Manipulated packages in the Python index gave attackers access to roughly 434,000 automated build pipelines across more than 2,500 companies, among them Airbus, Samsung, Salesforce and Volkswagen. Cloud keys, repository tokens, SSH keys, Kubernetes secrets and AI provider keys were taken. The malicious packages were online for only 40 minutes.
Source: heise.deRead article
Data Breach Costs·helpnetsecurity.comData breach cost 2026 averaged $4.99 million, AI attacks ran higher
The 2026 IBM Cost of a Data Breach Report, released this morning and based on Ponemon Institute interviews at more than 600 breached organizations, puts the average breach at $4.99 million — a record, and up more than a tenth year over year. Employees using unapproved AI tools figured in 43% of security incidents, more than double the prior year's share; about half of those incidents ended in data loss or compromise and roughly one in five drew a regulatory fine. Among organizations that suffered an incident involving an AI model or application, 92% lacked role-based access, multifactor authentication or comparable controls on those systems, and close to seven in ten breached organizations h
Source: helpnetsecurity.comRead article
Discover AI-Z
