skyseatTrust center
What we do today to protect your company’s data, what’s coming, with no fine print. And what we don’t have yet.
- Your data, in the EU: today, in the EU (the platform and its database in Helsinki, Finland; your AI employees’ computers in Germany, in pilot). If you need another country or region, such as the US, we can host it there on request.
- Every company kept apart: no one sees another company’s data, and automated tests check it.
- One computer per AI employee (in pilot): never shared with anyone.
- Credentials in a vault: the employee’s computer never sees them.
- An Activity log that can’t be tampered with: download it signed, and an auditor can check it without relying on us.
- A person in charge: they approve what matters, two people approve the riskiest actions, and 3 safeguards are always on.
- Encryption and backups: in transit, in the vault, in the Activity log signatures, and in backups kept off the server, with a disaster drill.
- Roadmap: SOC 2 Type I and an external firm’s pentest, with status.
Download the one-page security sheet (PDF) to forward to your IT or security team. Also in Spanish and Portuguese.
Where your data lives
The platform (the dashboard, the database, and the vault) runs on a Hetzner server in Helsinki, Finland. Your AI employees’ computers, in pilot, run on Hetzner servers in Germany. We back up every day, encrypted and stored outside the server too (see Encryption and backups). Everything between your browser and the platform is always encrypted (HTTPS).
Need your data in another country or region? The EU is the default location. If your company needs it hosted in another country or region (for example, the US), we can do that on request: we agree it with you before we start and put it in writing. Tell us what you need when you book a demo.
To think, your AI employee uses an AI model (Claude, from Anthropic) based in the US. It only gets what each task needs, and we use Anthropic’s commercial API, which doesn’t train its models on what it receives. We don’t use your data to train any model either.
This website is hosted on Vercel and gets no data from the platform or your AI employees. It has no cookies and only anonymous first-party analytics (Umami, on our own server).
Every company kept apart
- Each company has its own employees, tasks, Inbox, memory, and Activity log. If anyone asks for something that belongs to another company, the platform answers as if it didn’t exist.
- An automated test goes through every page and route of the dashboard trying to read and change another company’s data. New routes are added to it automatically.
- If you’re an agency, each client is a separate company: a client can’t see the agency or its other clients.
- People have a role (admin, manager, or view only), and view-only people can’t change anything.
- Single sign-on (SSO via OIDC or SAML 2.0): your company connects its own provider (Okta, Microsoft Entra ID, Google Workspace, ADFS…), verifies its domains through DNS, and can make it mandatory for its people. Anyone removed from your provider can no longer sign in, and with automatic provisioning (SCIM 2.0) they also lose access, sessions, and API keys immediately.
- Your admins choose how long a dashboard session lasts (from 8 hours), can sign everyone out at once, and see and revoke every API key that reaches the company. Every sign-in goes into your Activity log.
- Everyone can sign in with two-step verification: a code from an app like Google Authenticator or 1Password, with recovery codes. Your company can require it for everyone, and the Skyseat team always uses it.
One computer per AI employee
In pilot, each AI employee works on their own cloud computer: a server just for them, not shared with other employees or other companies. Their files and their browser stay there.
Their computer connects to the platform with its own access, which only works for their tasks and their company. If you let them go, that access is revoked instantly. And when they have no work, their computer shuts down.
Credentials in a vault
- Passwords and keys for the services you connect (Microsoft 365, Google Workspace, your invoicing software, your online store…) are stored encrypted (AES-256-GCM) in your company’s vault. The key that opens it is kept outside the database.
- The employee’s computer never sees them. When they need something from those services, they ask the platform, which checks they have permission, makes the call, and records it in the Activity log.
- If an AI employee creates an account on a service, the password goes into the vault. It never shows up in their conversation or in the Activity log.
- People’s passwords are stored with scrypt: not even we can read them. No one picks a password for someone else: new people get a single-use link.
An Activity log that can’t be erased
Everything your AI employees and your people do goes into the Activity log: what was done, when, with which account, and who approved it. The database doesn’t allow its rows to be changed or deleted by hand. The only thing that removes them is the Activity log retention period, if your company turns it on (2 years minimum; it doesn’t expire by default), and every purge is recorded.
You can download the Activity log for any period as a PDF, CSV, or JSON file for an audit or an inspection. It’s signed (Ed25519) and its rows are chained by their hashes: if anyone changes, removes, or adds a row, it shows. And you don’t have to take our word for it: an auditor can check it with a standalone single-file script, without skyseat and using our public key, which is published. It proves the file hasn’t been touched since you downloaded it. It helps you show what was done and who supervised it; on its own, it isn’t proof that you comply with the law.
A person in charge
You decide how much freedom each AI employee has: Supervised (asks before everything), Automatic (asks before sensitive actions), or Full autonomy (doesn’t ask; reports what they do). Whatever needs your OK reaches your Inbox and your phone, and, in pilot, you can watch their screen or take control at any time.
At every level, Full autonomy included, 3 safeguards are always on:
- Money. A monthly spending cap and a per-action cap: anything over it waits for a person. Plus your plan’s limits.
- Law. Anyone who talks to your AI employee by phone, email, or in a meeting knows they’re talking to an AI, and their instructions include GDPR.
- Activity log. Everything is recorded, including what was approved automatically.
And there are things they never do: get past a “prove you’re human” check (they hand it to you) or issue refunds or charges in your online store. Anything asked of them by phone always waits for a person to approve it.
Four-eyes approval
For the riskiest actions (payments above an amount, deleting, sending to many people, or changing permissions) you can require, per action and per role, one more person to approve. It waits in the Inbox for two different people, neither of them the one who assigned the task: it’s never approved automatically (not even in Full autonomy) or in bulk, and one person is enough to reject. Every signature goes into the Activity log.
Encryption and backups
| Between your browser and the platform | Always encrypted (HTTPS). |
|---|---|
| Credentials for the services you connect | Encrypted with AES-256-GCM in your company’s vault. The key to open it is kept outside the database. |
| People’s passwords | Stored with scrypt: not even we can read them. |
| Downloadable Activity log | Signed (Ed25519), with rows chained by their fingerprints. |
| Phone notifications | End-to-end encrypted. |
| Backups | One every day, plus one before each update. They leave the server encrypted (AES-256) to Hetzner storage in a different data center (Falkenstein, Germany), which sees nothing in the clear. We keep 14 daily and 8 weekly backups. |
| Tested restores | Every week the latest backup is restored into a temporary database and checked. On 10/31/2026 we ran a disaster drill: with the server lost and only what’s stored outside it, the platform was back up with its data in 4 min 18 s. It can be repeated whenever needed. If the server were lost, you could lose up to 24 hours of data. |
Your data is yours
- You can download all your company’s data from the dashboard, and each person can download their own.
- If you close your company’s account, everything is deleted after 30 days: employees and their computers, tasks, memory, connections, calls, and the Activity log. Until then, you can change your mind.
- You choose how long we keep calls, emails, chat conversations, attachments, and the Activity log (Account › Data and privacy). By default, calls and emails are kept 90 days and the rest doesn’t expire; if you shorten the Activity log, never below 2 years. Every purge is recorded in the Activity log. The voice provider deletes each call as soon as we’ve saved it.
- Everything an AI employee writes or says to an outside person says it’s an AI: email, WhatsApp, Slack, support chat, calls, meetings, calendar invitations, HubSpot, Linear, and Notion. The platform adds it, in the recipient’s language, and your company can reword the text but not remove it.
- We sign a data processing agreement with you: you decide what the data is used for, and we only process it to provide the service.
Who else is involved
These providers process data to deliver the service, under contract and only for that. We give you 30 days’ notice in the dashboard before adding or changing one.
| Hetzner | Servers for the platform and the computers, and backups. In the EU (Finland and Germany). |
|---|---|
| Anthropic | The AI model (Claude): gets what the employee needs for each task. In the US. |
| ElevenLabs | Voice and call transcription, only if you turn on the phone or voice. In the US. |
| Twilio | Phone numbers and lines, only if you turn on the phone. In the US. |
| Better Stack | Technical server logs and uptime monitoring. They may include IP addresses. In the US. |
| Browser notifications | They deliver notifications to your phone, only if you turn them on. The content is end-to-end encrypted: they only see which device it goes to. |
Anything that goes to the US is covered by GDPR safeguards. Every provider, with its country and safeguard, is listed under subprocessors. The services you connect yourself (Google, Microsoft, Slack…) aren’t our providers: they’re yours, under their own terms.
Security roadmap
What we want to achieve. There are no confirmed dates yet: we haven’t hired an outside auditor or provider for any of these steps. When there are, we’ll publish them here.
| Status | Target | |
|---|---|---|
| Legal documents reviewed by legal advisors (privacy policy, terms, and data processing agreement) | In progress | Before opening to customers |
| Light penetration test from the outside, run by our own team (10/31/2026): no exploitable flaws without an account, minor gaps noted | Done | First round |
| Penetration test (pentest) by an external firm | Planned | Date to be confirmed |
| SSO with your identity provider via OIDC (Okta, Microsoft Entra ID, Google Workspace, and other compatible providers) | Done | Available (10/31/2026) |
| SSO via SAML | Done | Available (10/31/2026) |
| SOC 2 Type I (an independent auditor’s report on the design of our controls) | Planned | Date to be confirmed |
| SOC 2 Type II (the controls working over an observation period) | After Type I | Date to be confirmed |
| ISO 27001 and ENS (if public-sector customers ask for it) | Not started | Date to be confirmed |
Until then, if your company needs a security questionnaire, we’ll fill it out with you: write to us from the dashboard or ask when you book a demo.
What we don’t have yet
- Certifications. We don’t have SOC 2, ISO 27001, or ENS yet. We review the platform’s security thoroughly before opening up and with every new feature, and we’ve run our own penetration test, but that isn’t the same as an audit or a pentest by an outside firm. The steps are in the roadmap.
- Legal documents in draft. The privacy policy, the terms, and the data processing agreement are being reviewed with legal advisors.
- In pilot, some things still in test. Employee computers, live screens, the phone, each employee’s own email, WhatsApp, and Microsoft 365 and Google Workspace still run in test mode. We’ll share it in What’s new when they’re live. What works today and what doesn’t, piece by piece.
Already a customer and have a question, or spotted something odd? Write to us from the dashboard, under Help and support: we’ll reply within 1 business day. And to see whether everything is working right now, check the status page.
Last reviewed: 11/1/2026.