Security

This page is for the people who approve Atlas for your company: security, IT and legal reviewers. It says what Atlas does, and just as clearly what is not done yet.

Your website stays in your control

• Atlas has no access to your source code, repository, servers, hosting account or deploy pipeline. Nothing in Atlas can deploy code.
• Your website pulls published content with a read-only delivery token, from server code. Atlas cannot push code or scripts onto your site.
• Content cannot run code. There are no plugins, HTML blocks or custom scripts. Editors fill typed fields your developers defined, and your components render them. This removes the most common way CMS-driven sites are compromised.
• Your site keeps a built-in fallback, so an Atlas outage does not take your website down.

Isolation between customers

• Every record belongs to an organization, site and environment. Atlas resolves that scope on the server for every request; it is never taken from the browser.
• The database enforces the same scope with row-level security. The application’s database role does not own tables and cannot bypass these rules.
• Media links and background jobs carry the same scope.
• Development and Production have separate content, tokens and settings.

Accounts and access

• People sign in through Atlas’s single sign-on service. Access is by invitation only.
• Roles limit what each person can do on each site (see Development and Production).
• Sensitive actions (approving a release, changing members, exporting data) can require a second confirmation on the person’s phone.
• Removing a person ends their sessions immediately.
• Every change is written to an audit log that cannot be edited. Organization admins can view and export it.

Tokens and secrets

• Delivery tokens, AI agent tokens and webhook secrets are shown once. Atlas stores only a fingerprint of tokens, so they cannot be read back, even by Atlas staff.
• Tokens are scoped to one environment and can be revoked at any time.
• Email passwords and API keys your site owner adds are encrypted, never shown again and never written to logs.
• Webhooks are signed (HMAC-SHA256 with a timestamp) and only go to public HTTPS addresses.

Production safety

• Production is off for every new site.
• Production serves only an approved, signed release. A release must pass its checks and be approved by a person with approval rights. AI agents can never approve or publish.
• Every release can be rolled back.

Abuse protection

• Forms and the chatbot use Cloudflare Turnstile, honeypots, timing checks, duplicate detection and rate limits.
• Browser endpoints accept requests only from your registered website origin.
• Page-view counts use no cookies and store no IP addresses or visitor identifiers.

AI

• AI agents connect over MCP with a token limited to one Development environment and the permissions you grant. They have no shell, database, file or web-fetch tools.
• The chatbot’s AI answers only from your site’s published content and is instructed to treat that content as data, never as instructions.

Not done yet

We would rather you hear this from us:

Item	Status
Independent penetration test	Not done
Certifications (SOC 2, ISO 27001)	None
Data residency choice	Not decided. Ask us before you sign
Data retention and deletion periods	Not decided
Single sign-on with your own identity provider (SAML/OIDC)	Not available
Uptime commitment (SLA)	None until the service has been measured in production
Point-in-time database recovery	Not available; daily backups are planned

Reporting a vulnerability

Email security@orbisai.ae. Please include the steps to reproduce and do not access other customers’ data. We will acknowledge your report and keep you informed. Our security.txt has the same details.