Security and privacy at MIDAS
This page explains where your data is stored, who can see it, and how MIDAS uses AI. Each technical claim links to its source.
*Last updated: November 16, 2026*
Four facts about your data
- Your data belongs to you. We do not sell your data. You can ask for a full copy of it at any time.
- We do not train AI on your data. Our AI provider's contract does not allow it. We train a custom model only if you sign a contract for it.
- Your data is stored in Singapore. Your database, files and backups are on Microsoft Azure, in its Singapore data center.
- Only your company can see it. Each person sees only what their role allows. Every change is saved with who made it and when.
What happens when you ask MIDAS AI
1. You ask MIDAS. In MIDAS chat on web, phone or Telegram.
2. The AI model answers. It runs privately in our own Microsoft Foundry account. By contract, it does not train on your data.
3. No training. Microsoft does not use the request or the answer to train AI.
You can also connect your own AI app, such as ChatGPT or Claude, to MIDAS. That app follows its own privacy rules, not ours. Check its settings for AI training.
Technical details
This part is for engineers and auditors. Each item names the exact service, setting or contract, so you can check it.
Hosting and data location
- Cloud provider: Microsoft Azure.
- Region: Southeast Asia (Singapore). The app, database, files and backups are all in this region.
- Database: Azure Database for PostgreSQL – Flexible Server.
- File storage: Azure Blob Storage. HTTPS only, TLS 1.2 or newer.
- Website: Azure Static Web Apps, delivered through Cloudflare.
AI processing
- AI service: AI models hosted on Microsoft Foundry, in NBR's own private Azure account. Other customers cannot reach it, and OpenAI never receives your data. (Source: Microsoft: data, privacy and security for Foundry models)
- Contract: Microsoft's handling of our data is governed by the Microsoft Products and Services Data Protection Addendum (DPA). (Source: Microsoft Products and Services DPA)
- Training: Microsoft's terms say requests and answers are not used to train or improve any AI model. They are not shared with OpenAI or other customers. (Source: Microsoft's terms (see the box at the top of the page))
- Chat history: MIDAS calls the AI with storage turned off (store = false). Microsoft keeps no chat history for us.
- Where AI runs: We use Microsoft's Global deployments. A request may be processed in any Microsoft region where the model runs. Your stored data stays in Singapore. (Source: Microsoft: location of processing for Global deployments)
- Safety review: Microsoft checks requests for abuse automatically. If a request is flagged, authorized Microsoft staff may review it. Flagged requests are not used for training. (Source: Microsoft: abuse monitoring)
- Backup provider: If Microsoft Foundry is down, MIDAS will switch to Amazon Bedrock so your service keeps running. AWS states that Bedrock does not store requests or answers, does not use them for training, and does not share them with model makers. (Source: AWS: Amazon Bedrock data protection)
- Rule for any new provider: Its contract must forbid training on your data. We add it to this page and tell customers at least 30 days before we use it.
Access control
- Company separation: Every record belongs to one company. MIDAS checks this on every request, so one company cannot read another company's data.
- Roles (least privilege): We follow the least-privilege principle: each person gets only the access their job needs, nothing more. A storekeeper who records stock moves does not see selling prices or customer debts. If an account is stolen or someone makes a mistake, the damage stays small.
- Change history: Orders, stock moves and approvals save who made each change and when.
- Passwords: Stored as bcrypt hashes. We cannot read your password.
- Connection keys: Keys for connected apps are encrypted before we save them.
- Outside AI apps (MCP): They connect with OAuth 2.1 and PKCE. An app gets only the access of the person who approved it.
- NBR staff: Only NBR engineers who run MIDAS can reach the live systems. We open your data only to help you, fix a problem or follow the law.
Encryption
- In transit: HTTPS with TLS 1.2 or newer. Data moving over the internet is scrambled so others cannot read it.
- At rest: Azure encrypts the database, files and backups with AES-256.
Backups and recovery
- Database backups: Automatic. Kept for 35 days.
- Restore: We can restore the database to any point in those 35 days.
- Backup location: The same Azure region as the database (Singapore).
Security incidents
- Notice: We tell you within 72 hours after we confirm an incident that affects your data.
- Data taken: If we confirm your data was taken, we tell you within 24 hours.
Keeping and deleting data
- Export: Ask us, and we export your data in a machine-readable format.
- Business data after you leave: Kept 90 days so you can export it, then deleted. Ask for a deletion certificate if you need one.
- Account details after you leave: Kept 30 days, then deleted.
- Billing records: Kept up to 7 years.
Subprocessors (companies that handle data for us)
- Microsoft Azure: Hosting, database, file storage, email and AI.
- Cloudflare: Delivers the website.
- Stripe: Card payments for MIDAS plans.
- Google: Only if you connect Google Calendar or Google Contacts.
- Telegram: Only if you connect a Telegram chat to MIDAS.
Common security questions
Do you use our data to train AI?
No. We do not train AI on your data. Microsoft, our AI provider, does not either. Its contract does not allow it. We train a custom model only if you ask and sign a contract for it.
Can OpenAI or Microsoft see our data?
OpenAI cannot. The models run at Microsoft, not at OpenAI. Microsoft checks requests for abuse automatically. Only flagged requests may be seen by authorized Microsoft staff.
Where is our data stored?
On Microsoft Azure, in Singapore. This includes the database, uploaded files and backups. AI requests may be processed in other Microsoft regions.
Do you have SOC 2 or ISO 27001?
MIDAS runs on Microsoft Azure. Azure holds SOC 1, SOC 2, SOC 3 and ISO/IEC 27001 certifications. You can download Microsoft's audit reports from the Microsoft Service Trust Portal.
Is our data encrypted?
Yes. Data moving over the internet uses HTTPS with TLS 1.2 or newer. Azure encrypts the stored database, files and backups with AES-256.
Who at NBR can see our data?
Only NBR engineers who run MIDAS can reach the live systems. We open your data only to help you, fix a problem or follow the law.
Can another company see our data?
No. Every record belongs to one company. MIDAS checks this on every request.
Do you back up our data?
Yes. The database is backed up automatically. Backups are kept for 35 days. We can restore the database to any point in that time.
What happens if there is a security incident?
We tell you within 72 hours after we confirm it. If your data was taken, we tell you within 24 hours. We share what we know and update you as we learn more.
Can we get a copy of our data, or delete it?
Yes. Ask us for an export in a machine-readable format. When you leave, you have 90 days to export. Then we delete your data. We can give you a deletion certificate.
Which AI providers do you use, and how will we know if this changes?
Today we use AI models hosted on Microsoft Foundry. Amazon Bedrock will be our backup, so MIDAS keeps working if Microsoft Foundry is down. Bedrock also does not train on your data. Any new provider's contract must forbid training on your data. We update this page and tell you at least 30 days before we use it.
Can you fill in our security questionnaire?
Yes. Email [email protected]. We can also send you our list of subprocessors and where they process data.
Security questions or a security report
Write to us about security questions, questionnaires or a possible security problem.