Home›Trust Center

Trust Center

Your risk committee
gets the dossier ready to go.

Most AI projects don't die on the technology: they die during security diligence, when the vendor has no document to hand over. Here's what Powerbizz puts on the table before you even ask.

Talk to our security team See security and governance ›

One agent actsanother audits the action before any execution

6 documentsdelivered to your risk team at the start of diligence

Already auditedby security committees at global insurers and hospital networks

Summary for the committee

Operating model: Managed AI Operations: Powerbizz operates, oversees and is accountable for the result
System access: Only via a secure API gateway, with minimum permission and no credential on the model
Data used for training: Contractually banned, including model subprocessors
Data officer: Named DPO, with a contact channel included in the documentation package

The principle

One agent acts.
Another audits.

Before writing anything to your system, a second, independent agent validates the action. When data is missing or confidence is low, it stops and escalates to a human. It never guesses.

Every agent has an owner, a scope and a version.Plus a kill switch that turns it off instantly.
Authority level defined by you.What the agent does on its own, and where it needs human approval, is written down.
Mandatory certification gate.Simulation with real cases and adversarial testing before going live. No critical agent skips this.
Knowledge restricted to your own.The agent answers from your organization's RAG, not the internet's generic knowledge.
Multi-model.We already run models from Anthropic, OpenAI, Google and Mistral, depending on the use case. Swapping the model behind it isn't a new project, and you're not locked into one AI vendor.

The path of an action

The agent proposes the actionReads the request, consults the organization's RAG and drafts the execution.

The independent auditor validates itKicks in when the agent extracts or writes data via API.

Low confidence stops and escalatesMissing data or ambiguity becomes a human task, never a guess.

Execution via a secure gatewayConnects to your ERP, CRM or core system without exposing the credential to the model.

Confirmation to the requesterResult returned to whoever asked, with what was executed.

Technical controls

What's already implemented.

No roadmap promises in this table: what's here is already running on production accounts.

Domain	Control	What it means for you
Data in transit and at rest	TLS 1.3 and AES-256	Interception and leaks via storage access fall outside any realistic scenario
Identity	MFA and corporate SSO (Okta, Microsoft Entra ID)	Access follows the identity policy your company already uses, with per-agent permissions (RBAC)
Personal data	PII tokenization	The model works without receiving identifiable data when it isn't needed
Client separation	Multi-tenant isolation	Your operation's knowledge and history never touch another client's
System access	Secure API gateway	Your ERP's credential never reaches the model, and every call is logged
Infrastructure	AWS, with multi-AZ failover	If one AWS zone goes down, the operation keeps running with no manual intervention
Availability	99.5% SLA, with a service credit	Downtime below the agreed level triggers a contractual credit, not just an apology
Critical incident response	Up to 4 hours	A contractual deadline for first response, not an internal goal with no teeth
Security testing	Annual pentest, aligned with SOC 2 and ISO	The control is verified by a third party, not just claimed by us
Use of your data	Contractual training ban	Your information never becomes anyone's model knowledge

Where the operation runs

Three ways to deploy,
the same governance in all three.

The committee decides how much of its own perimeter it needs. Powerbizz fits into it, without cutting back on control.

Shared managed cloud

Powerbizz operates the infrastructure on AWS, with logical isolation per client. The fastest path to getting the first agent into production.

Dedicated private VPC

A segregated network environment just for your operation, inside the same managed account. For the committee that requires physical isolation, not just logical.

On-premises or air-gapped

The operation runs inside your own perimeter, with data never leaving your network. For the client whose data can't, by policy or regulation, touch a third party's cloud.

Shared responsibility

What Powerbizz operates,
what stays with you.

The same model AWS uses to split responsibility between cloud and client, applied to the agent layer.

Who answers for what

Infrastructure, availability and platform patching: Powerbizz
AI model, agent orchestration and per-action auditing: Powerbizz
Gateway to your systems and the call trail: Powerbizz
Policy for who can access what inside your company: Client
Quality and freshness of source data, in your ERP, CRM or patient record: Client
Each agent's authority level, what it does alone and where it needs approval: Defined together, before the pilot

Documentation package

Six documents,
delivered before you ask.

Your risk team gets the documentation ready to go, not a blank questionnaire for us to fill out over two weeks.

Platform dossierArchitecture, data flow, technical controls and multi-tenant design. The document that answers most of your committee's questionnaire in one go.

ROPA inventoryRecord of processing activities, with purpose, legal basis, data categories and retention period per flow.

AI agent DPIAData protection impact assessment specific to agent-based operations, exactly where most vendors have no answer.

Incident response planSeverity classification, first-response time, escalation path and communication to the client's data officer.

Privacy and Security by DesignHow privacy and security enter the development cycle, before the code, not at final review.

Subprocessor listWho takes part in processing, in what role and under which contract clause, including model providers.

Request the full dossier Talk to our sales team ›

Privacy and data protection

Privacy treated as design,
not as a clause.

Operating with agents creates new data processing, with a new purpose. That's why the inventory and impact report are redone with every flow that goes into production, not once at contract signing.

Purpose and legal basis per flow, recorded in the ROPA inventory delivered to your data officer.
A DPIA specific to AI agents, which is exactly where most vendors have nothing to show.
Data subject rightsoperationalized: access, correction and deletion with a defined path and agreed deadline.
Retention and disposalwith a contractual deadline, and proof of disposal at contract end.
Sensitive data with its own design, in health and in benefits, built together with the client's legal team.

Questions your data officer asks

What's the legal basis?: Defined per flow and recorded, never assumed
How long do you keep it?: Contractual deadline, with proven disposal
Who else processes the data?: Subprocessor list, with role and clause
How do I handle a data subject request?: Path and deadline defined before the pilot
What if there's a breach?: Incident plan with an SLA and notification to the data officer

Keep reading

How governance shows up in each line.

Digital WorkforceAI agents that sell and support customers on WhatsApp and other channelsView page ›Power OpsRepetitive processes executed automatically, start to finishView page ›Power HumanAutomatic record of the team's visits and conversations, with human confirmationView page ›

Frequently asked questions

Your committee's questionnaire, answered.

Do you use our data to train models?

No, and that's in the contract, not just in a policy. The ban is explicit and covers the model subprocessors the platform uses.

Where is the data hosted?

In managed cloud on AWS, with per-client isolation and automatic failover across availability zones. For clients who require their own perimeter, a dedicated private VPC model and an on-premises model are also available. Hosting region, retention period and disposal process are defined in the contract and described in the platform dossier your team receives.

Are you SOC 2 and ISO 27001 certified?

The platform is built aligned with those controls, with an annual pentest and the corresponding documentation package. When your committee requires a formal third-party certificate, that's addressed explicitly during diligence, with no runaround.

Does the AI model have access to our systems' credentials?

Never. The credential stays in PowerBizzOS's secure API gateway. The agent requests the action, the gateway executes it with the minimum permission you defined, and returns the result, with every call logged.

How do you prevent the agent from making up an answer?

The agent responds based on your organization's own knowledge, an independent auditor agent validates the action against business rules, contract terms and authority level before any execution, and low confidence becomes a human task. On top of that, no critical agent goes into production without passing the certification gate, including adversarial testing.

Who is the data officer and how do we reach them?

There's a named DPO, and their contact channel is part of the documentation package delivered at the start of diligence, along with the incident response plan.

What happens if we want to end the contract?

Data is returned in the agreed format, with proven disposal, within the contractual deadline. The exit design is discussed before entry, not after.

Next step

Bring your company's
security questionnaire.

We answer with documents, not promises. If something your committee requires is missing, we tell you what exists and what doesn't, no runaround.

Schedule an executive diagnostic

Talk to BellaDigital attendant, replies right away

One agent acts.Another audits.