In the fast-paced world of artificial intelligence, trust is the most important currency. Companies implementing AI solutions must be able to rely on their sensitive data being protected at all times. This article was prompted by a security incident at the Austrian AI provider Localmind in October 2025, reported by heise online. It shows that a security architecture covers not only production systems but also test, demo and beta environments.
We do not want to pass judgement on the incident. Instead, we take it as an occasion to set out objectively which questions companies should ask when choosing an AI platform – and how we at innFactory answer them with our CompanyGPT.
The Occasion: A Misconfigured Test Instance
According to Localmind, a publicly reachable test instance of its own cloud infrastructure was misconfigured, and this gap was exploited from outside. heise online reported on the incident on 7 October 2025. According to its own account, Localmind took all services offline within 30 minutes and investigated the incident together with cybersecurity experts and IT forensics specialists.
On the outcome, Localmind states on its website (translated from German): “Data exfiltration cannot be proven; our customers’ on-premise systems were not affected.” According to Localmind, the reviews by the data protection authorities were closed without further measures. Since December 2025, the company says it no longer operates its own servers; the platform runs on-premise at customers’ sites or in certified partner data centres.
Regardless of the individual case, two principles apply to every AI platform: test, demo and beta environments belong in the security architecture just as much as production systems and should be separated from production data and credentials. And access rights – including those of newly created accounts – should be granted on the principle of least privilege.
The innFactory Approach: Security as a Design Principle of CompanyGPT
In developing CompanyGPT, we understood security not as an add-on feature but as the foundation of the entire architecture. Our approach rests on three principles:
1. Transparency Through Open Source We deliberately rely on proven open-source components. CompanyGPT is based on a fork of LibreChat, a widely used open-source solution that is developed further by an international community of several hundred contributors. According to its CEO, Shopify also uses this base.
- The Advantage: The source code of the platform base is publicly viewable. Potential vulnerabilities can be reviewed, reported and fixed by the community. This transparency creates verifiable trust.
2. Data Sovereignty Through Separate Tenants A core piece of our security strategy is the customer’s data sovereignty in their own cloud environment, for example in Microsoft Azure. With us, every CompanyGPT instance is a separate tenant, isolated from other customers.
- The Advantage: There are no shared databases, computing resources or configurations between customers. As a result, an incident at one customer cannot spread to other customers via shared databases or computing resources. Your data remains in your dedicated environment.
3. Seamless Access Control Through Established Enterprise Systems We don’t force new, separate user management systems. Instead, CompanyGPT integrates seamlessly into your existing IT infrastructure. Access is controlled via proven systems like Microsoft Entra ID (formerly Azure AD).
- The Advantage: Your IT department maintains full control. You can manage access rights granularly, revoke access for employees, or terminate sessions centrally – all with the tools they already know and trust. This also applies to custom connected MCP servers, which we can also secure via OAuth if needed.
Why Microsoft Azure? Security at Platform Level
Our decision for Microsoft Azure as a technological base is a central pillar of our security concept. Azure provides extensive security features and certifications on which CompanyGPT builds.
- GDPR-Compliant Data Storage: Data is stored in European data centres; with the right region and model deployment type (such as an EU data zone), processing also stays within the EU.
- Encryption: Data is encrypted in transit and at rest.
- Identity and Access Management: Azure offers tools for monitoring and controlling access. Actions can be logged, and unusual activities can be automatically detected and blocked.
- Zero Trust: Azure supports building a zero-trust architecture following the “never trust, always verify” principle: requests are authenticated and authorised instead of trusting access within the network by default.
What Companies Should Look For – and How CompanyGPT Implements It
| Criterion | What companies should look for | Implementation in CompanyGPT |
|---|---|---|
| Software Base | Is it clear which software the platform is based on and how security updates are applied? | Fork of the open-source platform LibreChat; source code of the base publicly viewable. |
| Architecture | Does the platform run in a separate environment per customer or in a shared environment, and how are customers separated from each other? | Separate instance per customer in the customer’s cloud subscription, with no shared databases or computing resources between customers. |
| Data Storage | Where are production and test data stored, and are test, demo and beta systems separated from production data? | Data in the customer’s cloud subscription; separation of production and test data. |
| Access Management | Can the platform be connected to the existing identity management, and are permissions granted on the principle of least privilege? | Integration with Microsoft Entra ID or Google; permissions and sessions are managed centrally by the customer’s IT. |
| Control & Audit | Is access logged, can the company view the logs itself, and are there regular security reviews? | Traceability through the open-source base and Azure security tools in the customer environment. |
Beyond that, these questions are worth asking any provider:
- Which test, demo and beta instances are publicly reachable, and how are they secured?
- What permissions do newly created accounts receive by default?
- Where are credentials and keys managed?
- Are there regular external penetration tests?
- How and how quickly does the provider inform customers about a security incident?
Conclusion: Trust Is Created Through Design
Security incidents can affect any provider. What matters is that security is anchored in the entire architecture – including test and beta systems and the way access rights are granted – and that companies can trace and verify this architecture.
With CompanyGPT, innFactory offers a solution that thinks about security from the ground up. The combination of transparency (open source), control (Azure & enterprise login), and isolation (dedicated tenants) creates a robust and resilient architecture. This ensures that our customers can use the benefits of AI without risking control over their most valuable asset: their data.