Security and trust
Security is built in, not bolted on.
Every MNB product and enterprise deployment is protected in two directions: who and what can reach your data and your AI, and whether what the AI says is backed by evidence. These are the controls we run today, in production, on our own products.
Two trust layers
Securing access is half the job. The other half is the answer.
Layer 1 · MNB Intelligence
The governed gateway between applications and AI models. One key per application, stored only as a hash. A key reaches only its own product's profiles. Failover between providers, recorded in a log that never keeps the prompt or the answer.
Layer 2 · MNB Veritas
The verification layer. Each factual claim in an AI answer is checked against approved sources with a verbatim quote, figures and financial years are compared in code, and written policy decides whether the answer is approved, qualified, held back or sent to a person.
Tested, and published
On 14 November 2026, AbroBot answered 30 of 30 official US Department of Education figures correctly; general models without web search scored 9 and 0. A small test run by MNB, with its limits stated.
Under both layers
The controls in every product.
Hardened hosting
Our products run on enterprise cloud platforms, Vercel and Google Cloud, with automatic DDoS mitigation and HTTPS on every request. Apps are placed in the same region as their data: Mumbai for our India-hosted products.
Customer isolation
Row-level security inside the database means each customer can only read and write its own records, enforced by the database itself, not only by application code.
Secrets that stay secret
Provider and payment keys live in encrypted server-side secrets, never in code or browsers. AuditFlow scans every change for leaked secrets before it ships.
AI from the server only
Our products call AI models from the server, never from the browser, so model keys are never exposed to users.
Verified integrations
AuditFlow's payment webhooks are signature-verified and protected against duplicate events. Scheduled jobs refuse to run without a secret. NyayaAI's database connections use fully verified TLS.
Safe releases
Every release is traceable to a specific code commit, previous versions stay ready for instant rollback, skew protection keeps users on a consistent version, and a hard spend cap prevents runaway costs.
Product by product
What each product protects, and how.
The same standard, applied to the data each product holds.
| Product | What it protects | Key controls |
|---|---|---|
| MNB Intelligence | AI traffic for enterprise deployments | Hashed app keys, per-key model allow-lists, rate limits, logs without prompt or answer text, Mumbai hosting |
| MNB Veritas | What an AI answer states as fact | Verbatim-quote evidence, figures compared in code, an examiner from a different AI lab, written policy, organisation-scoped records |
| AuditFlow | Audit evidence and findings | Row-level security, signed payment webhooks, secret scanning on every change, Mumbai hosting |
| MNB Cortex | Business and finance data | Row-level security, secret-protected scheduled jobs, Mumbai hosting |
| NyayaAI | Legal documents and matters | Documents encrypted with a separate key-encryption key, verified-TLS database connections |
| TaxSense AI | Income and tax records | Row-level security, server-side AI access only |
| AbroBot | Student profiles and documents | Prompt-injection screening, answers from official records |
Controls as they run in production in November 2026. Veritas is rolled out product by product, starting where the cost of an error is highest. Enterprise deployments can add your own identity provider, network rules and logging.
Built for the DPDP Act
- Collect only what a feature needs
- Store data in India where the product's database is in India
- Access limited by role and by customer
- A named grievance officer for data requests
- Written details for your own DPDP assessment
What we will not claim
- SOC 2, ISO 27001 or GDPR certification
- An uptime SLA
- That any AI is right every time
- That every processing step stays in India on shared AI providers
Trust is earned in detail.
Ask for the security pack: an architecture overview, every control on this page in detail, and written answers to your security questionnaire.
Questions
Frequently asked.
How does MNB secure AI, not just data?
With two layers. MNB Intelligence governs access: who may call which model, with hashed keys, scoped profiles and logs. MNB Veritas governs answers: every factual claim is checked against approved sources before it is presented as fact.
Can one customer see another customer's data?
No. Products such as AuditFlow, MNB Cortex and TaxSense AI enforce row-level security inside the database, so every query is limited to the signed-in customer's own records.
Where is customer data hosted?
Each product's database region is chosen per product, and apps run in the same region as their data. AuditFlow and YarnTally run in Mumbai next to their databases in India; MNB Cortex and the MNB Intelligence gateway also run in Mumbai. Enterprise deployments can run on your own cloud account or servers.
Is MNB ready for the DPDP Act?
Our products are designed around DPDP Act principles: data kept to what a feature needs, stored in India where the product's database is in India, protected by access controls, and handled through a named grievance officer. We provide details in writing for your own assessment.
Is MNB certified?
Not yet. MNB does not claim SOC 2, ISO 27001 or GDPR certification. We share our controls in writing and walk your team through them instead.