Security

Built to be trusted with serious work.

We deploy AI into high-stakes operational work, so security, privacy and control are designed in from the start, not added on later.

Talk to us about security
Security by design
Control is built into how the systems work.

Security is not a layer we add at the end. It shapes how every system is built, deployed and run.

Dedicated environments

Client work runs in isolated, access-controlled environments, separated from other clients.

Least-privilege access

People and systems see only what they need, with permissions and oversight throughout.

No training on your data

Your documents and data are never used to train shared or third-party models.

Human in the loop

A named reviewer signs off. The system supports judgement; it does not replace it.

Source traceability

Every figure and claim traces back to its source document, so outputs can be checked.

Audit and logging

Every step, input and change is recorded and reviewable after the fact.

Data and privacy
Your data stays yours.

We handle company and personal data with care, and design deployments so you keep control of it.

Data minimisation
We work with the data a task needs, and no more.
Personal data
Personal data is handled with care, limited in use, and removed where it is not required.
Retention and deletion
Data is kept only as long as the work requires, and deleted on request.
Confidentiality
Client information is treated as confidential and never shared between clients.
Residency
We can keep data within agreed environments and locations where that is required.
Access and oversight
Access is controlled, logged and reviewable, with permissions matched to role.
Guardrails and human review
The system shows its work, and people stay in control.

Outputs are supervised, explained and checked. The system makes people faster and more consistent; it does not act on its own.

Reasoning shown

Outputs come with the reasoning and evidence behind them, not just an answer.

Confidence levels

Every output carries a confidence level, so reviewers know where to look first.

Escalation thresholds

Items beyond a threshold or below a confidence level route to a named reviewer.

Nothing silent

Nothing of consequence happens without a person in the loop.

Computed, not generated

Numbers run through tested, version-controlled methods, so figures stay consistent and checkable, not invented by a model.

Governed overrides

Any override of a check is recorded, with who authorised it and why.

Working with your security team
We fit your environment, and your process.

We deploy inside your environment where needed, work alongside your IT and security teams, and support your due diligence and security questionnaires. The aim is to make adoption safe and straightforward, not to work around your controls.

Start here

Have a security question? Ask us directly.