Safety and privacy
We take safety and privacy very seriously. Our AI assistants comply with the law, including the AI Act and the GDPR. They are secured technically in several layers.
Our approach
An AI assistant of ours has two parts: the chat window on your website and the software on our servers.
Chat window
The chat window runs on the website where it is placed. It is loaded with one piece of code and runs in an isolated Shadow DOM. This keeps the assistant fully separate from the rest of the website. Styles and scripts do not get in each other's way, and the assistant cannot reach sensitive parts of the page.
Software
The software of the assistant itself runs on Google Cloud Platform in Europe (region europe-west4). Answers come from AI models that work with checked sources of knowledge. That makes them as accurate as possible. The organisation chooses which model is used, and can switch easily.
Laws and regulations
In the European Union, the AI Act (Regulation (EU) 2024/1689) governs how AI is used. The law aims to use the opportunities of AI and to protect people's health, safety and fundamental rights. That is why the AI Act sorts AI systems by their risk.
There are four risk categories:
Our AI assistants fall within the limited risk category, because they mainly inform and explain. They answer from checked sources and say that they are AI. They take no decisions with direct medical or legal consequences, and have no effect on fundamental rights.
Our measures
This is how we make sure our AI assistants comply with the AI Act and the GDPR:
Transparency
The assistant tells visitors that they are talking to AI, as Article 50 of the AI Act requires. The chat window's menu has a link to the privacy statement. It says what happens to the messages.
Data protection
The assistant processes personal data in line with the GDPR (EU 2016/679). We keep it confidential and secure. The assistant does not ask for personal data and does not link conversations to an account. If a visitor does share personal data, Google Cloud Sensitive Data Protection (DLP) removes it automatically from the stored messages.
Data processing agreement
When an organisation has another party process personal data, the GDPR requires a data processing agreement. A message is personal data as soon as a visitor says something about themselves. That is why we sign a data processing agreement with every organisation.
Automatic deletion of data
Conversations are deleted automatically after a retention period set for each assistant. For Lampie and most other assistants it is 90 days. A few assistants, such as Doorloopjes, keep conversations without an end date by choice. The database deletes conversations itself, with a TTL (time-to-live). A daily clean-up checks it once more. On request we also delete conversations at once (the right to be forgotten).
Technical security
Our platform is secured in several layers, from the servers to the software itself.
Encrypted transport (HTTPS/TLS)
Everything between the chat window and our servers goes over encrypted HTTPS connections. A strict HSTS policy (HTTP Strict Transport Security) allows no unencrypted connection.
Isolated service architecture
Each part of our platform is a service of its own. It has its own service account and as few permissions as possible (least privilege). Tools and sources of knowledge can only be reached internally, not from the internet.
Security headers
All services send security headers, including Content Security Policy (CSP), X-Content-Type-Options and Referrer-Policy. These protect against common attacks, such as cross-site scripting (XSS) and clickjacking.
Prompt injection prevention
What a visitor types is checked automatically for known injection patterns and made harmless. We also watch what the AI model answers. That way we notice if the model gives away its instructions or repeats malicious input.
Rate limiting
The platform limits the number of requests per visitor. So no one can overload or misuse the platform at the expense of other visitors.
Ownership validation
Each conversation belongs to one assistant. With every request, the platform checks that the conversation belongs to that assistant. So no assistant can reach the conversations of another.
Secrets management
Keys and passwords are kept in Google Cloud Secret Manager. They are never visible in the code or in configuration files.
Removing personal data
Visitors may type personal data by mistake: a name, email address, phone number, BSN (citizen service number) or IBAN. Google Cloud Sensitive Data Protection (DLP) finds and removes it automatically, before anything is stored.
Monitoring and logging
All services write structured logs, with tracing through OpenTelemetry. Attempts at injection or at access without permission are logged. We look into them and follow them up.
Infrastructure
Google Cloud Platform
Our platform runs entirely on Google Cloud Platform in Europe (europe-west4, the Netherlands). All storage and servers are within the EU. The answers come from the OpenAI API, with a data processing agreement (DPA). Under OpenAI's policy, the data is not used to train models. We use managed services such as Cloud Run, Firestore and Cloud Storage, with Google's highest security standards.
Our own infrastructure
All assistants run on our own cloud infrastructure, managed as Infrastructure as Code (Terraform). This keeps the configuration, the security and compliance with the rules fully in our own hands.
Checked knowledge base
The knowledge base behind each assistant is well ordered, checked and updated regularly. The answers come from these sources, not from random places on the internet.
Continuous monitoring
We check the conversations systematically. In a dashboard we follow conversations and feedback, so we keep making the assistants better and safer.
Questions or more information?
Do you have a question about our technology or about how we comply with the rules? Please get in touch.
Get in touch