Security & safety
How we keep your work safe.
Plain answers about your data, who can reach it, what our AI agents may do, and what happens when something goes wrong. Where something depends on your project, we say so and confirm it in writing.
Your data stays where you choose: on your own servers, in your own cloud account, or with a contracted model provider. We confirm which in writing.
Six commitments.
These are promises. Your signed agreement sets the exact terms.
- A person approves anything that can't be undone
- We agree the approval points with you before go-live.
- Every action is logged
- Who did what, when, and from which source. The log is readable and can be exported.
- Your data stays where you choose
- Your own servers, your own cloud account, or a contracted model provider, stated in writing. Your data is never used to train shared models. If fine-tuning helps, we do it only with your written approval, inside your environment.
- Agents say they are agents
- No agent pretends to be a person, in any language.
- When unsure, they stop and ask
- Anything below the line goes to a person, with the reason attached.
- You can switch them off
- Every agent has an off switch you control.
Your data and access.
- Access
- Only named people on our team can reach your systems and data. You hold the list, and access ends when the work does.
- Connections
- Connections to our systems are encrypted.
Never decided by an agent alone.
Whatever its confidence, an agent prepares these and a person decides.
- Moving money or approving payments.
- Credit, lending or insurance decisions about a person.
- Hiring, firing or judging someone's performance.
- Medical or legal advice to the public.
What stays with a person, by industry.
- Travel & tourism
- Any price the booking system doesn't have
- Banking & finance
- Any account action or credit decision
- Insurance
- Every claim decision and payout
- Retail & e-commerce
- Refunds and discounts
- Healthcare & pharmacy
- Anything medical goes to a pharmacist or a doctor
- Hospitality & food
- Complaints and compensation
- Real estate
- Prices, negotiation and contracts
When an agent is unsure.
Pick how sure the agent is and watch what it does. When it is unsure, it stops and names who decides.
Confidence
Requests this morning, example data
How an agent shows doubt.
A source for every answer
The document and page, or the system and record, it came from.
Unknowns said plainly
“I couldn't read the date on page 3.” It says what it couldn't read instead of guessing.
Who decides
When it stops, it names the person who decides and sends the case with its notes.
Before any agent goes live.
- 1
Agree where it runs
Your own servers, your own cloud account or a contracted model provider, written into the proposal.
- 2
Agree who has access
The named people on our side and on yours, and what each of them can reach.
- 3
Agree the approval points
What the agent may do on its own, and what waits for a person.
- 4
Open the log to your team
Your team can read every action, with its source, as soon as the agent goes live.
If something goes wrong.
The more serious it is, the calmer we get: facts, impact and actions.
- 1
Stop the step
We pause the agent or the step that went wrong. Your team can carry on by hand.
- 2
Tell your named owner
The person you named on your side hears from us: what happened, and what it affected.
- 3
Explain from the log
What the agent did, when, and from which source.
- 4
Keep you updated
At the times we give you, until it's closed.
- 5
Review
A short written review: the cause, the fix, and what stops it happening again.
You can switch any agent off, at any time.
Questions about security.
Is our data used to train AI models?
Your data is never used to train shared models. If fine-tuning helps, we do it only with your written approval, inside your environment.
Which AI model do you use?
It depends on the delivery option and the work. We name the exact model, and where it runs, in your proposal and contract.
Can our security team review the setup?
Yes. Before go-live, we share the architecture, the delivery option, the access list and a sample of the log.
What happens to our data when the work ends?
We agree it in writing: what is returned to you, what is deleted, and when.
Bring your security questions.
Send us your security questionnaire or your list of controls, and we'll go through it with you point by point.
Write to usYou decide. We answer for it.