Data protection and security
How we protect the files and data you give us
When you send us a credit file, or your officers work with Mwikila, our AI Senior Banking & Finance Manager, you give us information about borrowers, your institution and your staff. Below we answer the questions a lender's risk team would raise, and describe each protection in turn.
Where something is not in place yet, the answer says so.
Questions to ask before you send us a file
If you look after risk or data protection at a lender, these are probably the first things you will want to know.
Who is responsible for protecting our data: you or the AI companies?
Both, in a chain the law recognises. Your institution stays the owner and controller of its data. LitFin Intelligence is your processor: we act only on your instructions, under a written data processing agreement. The AI providers we use are our sub-processors: they may use your data only to answer the request. For most of them that limit is written into a signed data processing agreement. For the inference providers added in August 2026 it is their published business-tier policy, which we have read but have not yet signed an agreement on, and our register names them.
This chain is public. Every outside service that touches data is on our sub-processor register, with the date it was added and whether we gave the notice period before it received data. We did not give it for the providers added on 7 August 2026, and the register says so.
Is our data used to train AI models?
No. We never train a shared model on your data, and the AI providers we use do not train on data sent through their business services. For our contracted providers that is a written term of their commercial agreement. For the inference providers added in August 2026 it is their published business-tier policy only: we have not signed a contract with them that says so.
We do not rewrite your content before sending it to our AI providers, and we chose that on purpose. We tried redacting it, and it damaged the documents our customers depend on: a bank statement with the lender names removed cannot be reconciled, and an officer's own company history came back with its awards and ministries replaced by placeholders. A document you send has to come back as you sent it. What protects you instead is the contracts and the way our software is built: most providers work under a data processing agreement with no-training terms, our sub-processor register (linked below) names the ones that do not yet, and what Mwikila learns about your staff stays locked to your institution and to each verified login. Our AI providers operate in several countries, including China, and the register names each one.
Are you built on one AI model, or many?
Many. We are not tied to any one AI model. If a provider fails or slows down, the request moves to the next one automatically, and a provider that keeps failing is taken out of use until it works properly again.
So one company's outage does not take Mwikila away from your team. Our voice and research services switch providers the same way.
What happens when a provider retires or blocks a model?
Providers retire models routinely, and the major ones publish retirement policies that give advance notice, in practice months ahead. Because LitFin Intelligence does not depend on any one model, a retirement means we move that work to a successor model; your service does not stop.
Every model we use can be swapped for another. The leading providers have also committed publicly to long notice periods, and some even to keeping retired models available.
What agreements does LitFin Intelligence have with its AI providers?
Commercial API agreements with data processing terms: the provider acts as a processor, may not use your data for its own purposes, does not train on it, and keeps it only briefly for abuse monitoring, within published retention limits.
Some of the providers whose infrastructure we use hold SOC 2 Type II, ISO 27001 and ISO 42001 certifications. We do not claim them for every provider, because we have not checked them for the providers added during 2026. Our register names every provider, and its purpose is to tell you before a new provider receives your data, so you have time to object. We did not do that for the providers added on 7 August 2026: we made a business decision to send them data first, and we have not yet issued the notice.
What are the chances of being hacked?
No company can promise you that the chance is zero. What we can show you is what an attacker would have to get past: each customer's data walled off by the database itself, encryption in transit and in storage with the most sensitive fields encrypted one by one, a role check on every request, a limit on how many requests anyone can make, rejection of forged requests, limits on where our servers can send data, and an audit log that shows any tampering.
The damage is also limited by design: LitFin Intelligence cannot move money and does not make lending decisions, so a breach here cannot empty an account or approve a loan. If a breach affecting personal data ever happens, Tanzanian law requires us to notify the Commission without undue delay, and we will.
What protects your data inside our software
Each institution's data is kept apart
Each institution's data sits behind rules built into the database itself. If a part of our software ever failed to filter a request, the database would still refuse to hand your data to anyone else.The most sensitive fields have their own encryption
Your connection to us is encrypted, and so is your data while it is stored. The most sensitive fields, such as ID numbers and sign-in credentials, are each encrypted with their own key before they reach the database, so a stolen record is of no use on its own.What Mwikila remembers stays with you
Mwikila adjusts to your role, your history and your files, so his help gets more useful the more you work with him. What he learns is read from and written to your own record only, tied to your verified login, and nothing you do is visible to another institution. Managers and administrators see only summaries with names removed. Your content reaches the AI model as you wrote it, names and identifiers included, and we never train a shared model on it.Each person sees only what their role allows
Before we release a new screen, an automated test checks that it asks for the person's role. Requests forged by a malicious website are rejected before they reach your account. And when Mwikila fetches research from the web, that traffic is kept away from our internal servers and cloud secrets, which closes one of the commonest ways attackers try to get behind a service.A log that shows any later change
Actions that matter for security are written to a chained log, so any later change to it can be detected, and the log itself avoids storing personal identifiers. We read files, appraise them and advise. Our software cannot move money, and the lending decision always stays with your people.A public list of who processes your data
We publish a list of every outside service that touches data, where it runs and what it can see. It is kept up to date, in a form software can read and check.
Your rights over your data
Tanzania's Personal Data Protection Act gives you the rights listed below. You can use most of them from this page, without writing to anyone.
- See everything we hold about you
- Take a copy in a file other software can read
- Correct anything that is wrong
- Ask us to delete it, keeping only what the law requires us to retain, with nothing left that links it to you
Each button below files your request with every system that holds your data, and gives you a reference number.
Things we do not claim
Each of these is a limit you should know about before you send us a file.
- We do not claim any security certification we have not been awarded. Our controls follow recognised security principles, and if we start a formal certification we will say so while it is under way and again when it is complete.
- We never call our software unbreakable, because nobody can promise that of any software.
- Your conversations are encrypted on the way to us and while they are stored, but not end to end: our servers have to read them to do the work, and they reach our AI providers as you wrote them. The answer on training above explains why.
- Some of the AI processing happens outside East Africa, so not all of your data stays in one country. Every place it is processed is listed on our data transfers page, under legal safeguards.
- Two-factor sign-in is not live yet. Today access is protected by sign-in details your institution issues, permissions set by role and session controls. We plan to add it, and we will tell you here when it is switched on.
- Tanzania's data protection law runs a registration and transfer-permit regime through the Personal Data Protection Commission. We track those obligations with our advisers. If you ask, we will tell you in writing where our registration and permits stand.
Tanzania's data protection law
Tanzania's Personal Data Protection Act (2022) and its 2023 regulations are in force and enforced. Below is what they require of a service like ours, and how we meet each requirement.
What the law requires
- Registration of data controllers and processors with the Personal Data Protection Commission, on a certificate renewed every five years
- An appointed data protection officer
- A written contract governing every processor relationship
- Security safeguards proportionate to the data, and breach notification to the Commission without undue delay
- Restricted cross-border transfers: personal data leaves Tanzania only under the Act's conditions and the Commission's permit regime
How we meet it
- A data processing agreement offered to every institution, with the controller and processor roles stated plainly
- Requests to see, export, object to or delete your data, sent from this page with a reference number
- A public register of every sub-processor, and a page setting out every cross-border transfer
- A data protection officer you can reach directly, and, for any institution that asks, a written account of where we stand on each obligation
If you are a lender, your regulator's rules on outsourcing and cloud services apply to your use of any outside service, ours included. We prepared our documentation, the separation of each customer's data and our commitment to tell you about a breach with that review in mind, and we will support your approval process with the Bank of Tanzania.
Write to our data protection officer at dpo@litfin-training.com with any question about your data.