Security and TrustData, AI, control and evidence

AI agents your CISO can approve

FluxForce puts AI agents inside your financial crime work. Before anything goes live, your CISO should be able to answer four questions: where the data goes, where the AI runs, who can make an agent act, and what record is left behind. This page answers all four.

Before anything goes live

What your security team will ask

Each question has a straight answer below. If you need more than this page gives you, we'll walk your team through the architecture under NDA.

Answered below
Data

Where do our data and the AI run?

SaaS, on-premise or hybrid, on Microsoft Azure. Rules and scoring decide what reaches an analyst. Nothing is migrated.

Answered below
Control

Who can make an agent act?

You set autonomy per agent and per risk band. High-risk work always goes to a person.

Answered below
Evidence

What record is left behind?

Every step is kept as tamper-evident evidence an examiner can replay.

The answers

How FluxForce handles data, AI, control and evidence

Written for the CISO and the CTO. Your MLRO will find the control and evidence answers here too.

ANSWER 01 / 06

Where FluxForce runs

SaaS, on-premise or hybrid, built on Microsoft Azure.
On SaaS, we host it in an Azure region agreed with you and map the data flow before you sign.
On hybrid, the agents run inside your environment and call services we host for specific steps. On-premise, the platform runs in your infrastructure.
Agents connect beside your core banking, screening and case management systems through APIs or scheduled file transfers. Nothing is migrated.
ANSWER 02 / 06

Where the AI runs

Deterministic rules run first, then a scored model. Rules and scoring decide what reaches an analyst.
A language model drafts and summarizes case notes and reads transactions against your written policy. Those steps run on Azure OpenAI today; on-premise, we run a model inside your environment.
Personal data is tokenized when it comes in, and card data is masked.
One customer's data never trains another customer's models.
ANSWER 03 / 06

Who can make an agent act

You set autonomy per agent and per risk band. Low-risk steps run on their own only if you switch that on after shadow mode.
Medium-risk work goes to a person by default. High-risk work always does.
Every agent has a kill switch. It turns the agent off at once, and the switch itself is recorded.
STRs, SARs and CTRs are drafted for your MLRO. Nothing is filed without that approval.
ANSWER 04 / 06

Access and sign-in

Your team signs in through Microsoft Entra ID (Azure AD) by default. Other identity providers are set up per customer.
Access is role-based, so each person sees and does only what their role allows.
Every session goes through zero trust checks.
Releases and container images are signed.
ANSWER 05 / 06

What record is left behind

Every agent step is written to tamper-evident evidence storage, so you can show whether a record was changed.
The record holds the alert that started it, the data the agent looked at, the settings in force, its recommendation and reasons, and your analyst's decision.
An examiner can replay any case in order.
Model versions are tracked and watched for drift, with rollback. The record is built for SR 11-7 and the EU AI Act.
ANSWER 06 / 06

Certifications and reviews

We don't hold SOC 2 or ISO 27001 certification today. We'll answer your security questionnaire in full, in your format or ours.
We'll walk your security team through the architecture under NDA and give you a data-flow diagram for your deployment model.
Our subprocessor is Microsoft (Azure and Azure OpenAI).
Our incident notification commitment is written into your contract, and on-premise customers get an exit path in the contract too.
Questions? We Have Answers

Frequently Asked Questions

That depends on the deployment model you choose. On SaaS we host in an Azure region agreed with you; on-premise and hybrid keep more of the processing in your environment. We map the data flow for your deployment before you sign.

It can hold a payment only if you set that risk band to allow it. High-risk work always goes to a person, and every agent has a kill switch. An agent never files a report. Your MLRO approves every filing.

Open the case. Every step is in the evidence record in order, with the inputs, the reasoning, the settings in force, the recommendation and the human decision.

The language-model steps run on Azure OpenAI today. Rules, scoring and sanctions matching don't need a language model. On-premise, we run a model inside your environment, confirmed at deployment design.

No. One customer's data never trains another customer's models.

We don't hold SOC 2 or ISO 27001 certification today. We'll answer your security questionnaire in full and walk your team through the architecture under NDA.

Take the first step

Request a security review

Send us your questionnaire or book an architecture walkthrough with your security team. We'll answer in full, under NDA if you need it.

Security reviewUNDER NDA
Security questionnaire
Architecture walkthrough
Data-flow diagram
Shadow mode on your dataSTEP 4