Skip to content
Early access: SafeHand is choosing its first design partners. Apply
SafeHand
Contact sales БГ Book a demo
A city skyline at dusk, reflected in a river
Security and trust

Security and sovereignty,
by design.

Your customers’ data stays where you need it. We start in EU regions, with private-cloud and on-premises options for the strictest requirements. Every agent action is permissioned, logged and reviewable.

Three ways to deploy.
One platform.

Choose where SafeHand runs. Agents, guardrails and evaluations work the same way in all three.

SafeHand CloudFastest start
Your private cloudFor regulated enterprises
On-premisesFor the strictest requirements
Availability
First, with design partners
On request, Google Cloud first
Planned, on request
Where it runs
EU regions first, operated by SafeHand
Your AWS, Google Cloud or Azure account
Your own data centre
Customer data
Stays in your agreed region
Stays in your account
Never leaves your network
Models
Hosted in your region
Hosted in your region, or your own
Self-hosted open-weight
Encryption keys
Managed by SafeHand
Your keys
Your keys
Operated by
SafeHand
SafeHand, inside your controls
Your team, with our engineers

Data protection.

Conversations with your customers are some of the most sensitive data you hold. We treat them that way, by default.

GDPR as standardA data processing agreement in every contract, data minimisation and support for data subject requests.
Data residencyProcessing and storage in the region we agree, EU by default, with no transfers you haven’t approved.
Encrypted everywhereTLS in transit and AES-256 at rest, for conversations, recordings and logs.
No training without your agreementCustomer data isn’t used to train shared models unless you agree to it in writing.
Role-based accessYour team sees what their role allows, and every access is logged.
Retention you setKeep recordings and transcripts as long as you need them, and no longer.

Responsible AI, built into every conversation.

Rules like the EU AI Act set clear expectations for AI that talks to people. Our agents are designed around them.

AI disclosureAgents tell people they are talking to AI at the start of every conversation.
A person on requestCustomers can always ask for a human, and get one with full context.
Guardrails and approvalsYou decide which actions an agent may take alone and which need sign-off.
Tested before and after launchEvery agent is evaluated on test conversations before it goes live, and continuously after.

Questions security teams ask.

Talk to us about security early. We share our architecture, data flows and subprocessors with design partners, and we’re preparing a full security pack.

Talk to us about security
Where is our data stored?

In the region we agree, EU by default. With private cloud or on-premises deployment, it stays in your own account or data centre.

Do you use our data to train AI models?

Not without your written agreement. By default, customer data is used only to run and improve your own agents, under your instructions.

Can SafeHand run inside our own cloud?

Yes, on request. We start with your Google Cloud project, with AWS and Azure when needed, under your security controls. On-premises with self-hosted models is planned; talk to us about timing.

How do agents hand over to a person?

Whenever a rule says so, when the customer asks, or when the agent isn’t confident. Your team receives a written summary, so nobody has to repeat themselves.

How do you approach the EU AI Act?

Agents always disclose that they are AI, every decision is logged and reviewable, and you control which actions need human approval.

Who can access recordings and transcripts?

Only the people you grant access to, by role. Our engineers access your data only when you ask them to, and every access is logged.

Two hands clasped together

Put your customer conversations in safe hands.

Bring one customer journey. In 30 minutes we’ll show you how a SafeHand agent would handle it, and what it would cost per outcome.