Sign in Start free trial
Platform

Security & Trust

Data isolation, permissions, approvals and audit built into the platform, not bolted on.

 AI enforces who can see and change what in the database itself, so every screen, report, export and AI answer follows the same rules. Sensitive changes need a second person, every important action is recorded, and AI can be limited or stopped. This page describes controls that exist in the product; it makes no claim about third-party certifications.

51 features available 7 planned 8 capability areas
Highlights

What stands out

Isolation in the database

Tenant separation is enforced where the data lives, so no screen can skip it.

Second person for sensitive changes

Makers cannot check their own work, including through delegation.

Audit you cannot rewrite

The audit log refuses updates and deletes from every role.

Stop AI in a moment

Kill switches halt agents and in-flight model calls.

Erase with a key

Destroying a person's key erases their sensitive data.

Capabilities

Every feature, in one list

Search or filter. Planned items are labelled; everything else is available now.

Tenant and row isolation

Each company's data is separated by the database, not just by application code.

6 features
  • Row-level security on business tablesTenant isolation is enabled and enforced in the database on the great majority of tables, with explicit grants and guard triggers.
  • Reports run as the viewerDashboards, exports and shared reports apply the viewer's own row and field security.
  • No blanket bypass for cross-tenant jobsCross-tenant dispatch uses narrowly granted functions instead of a role that bypasses isolation.
  • Isolation tests in the buildAutomated tests check tenant isolation, function allow-lists and table scopes.
  • Masked refresh into testProduction data copied to test is masked by default and sensitive classes are never copied.
  • Non-production safeguardsTest tenants cannot send real e-mail, messages or payments.

Permissions and access

Fine-grained rights per hub, role, field and record.

11 features
  • Roles and permission matrixRoles grant actions per hub and sub-module with scope and own-only variants.
  • Per-hub seatsNo seat, no access in production tenants, with seat types and pools per hub.
  • Field-level securityFields can be hidden or read-only per role; hidden means absent everywhere, including AI and exports.
  • Approval limits per roleMaximum amounts a role may approve, per currency.
  • Sharing rules and record grantsOpen specific records or subtrees to specific people, with expiry and a reason.
  • Explain an access decisionAsk why a person can or cannot do something.
  • Delegated administrationAdmins can grant only what they are allowed to, enforced in the database.
  • Directory syncProvision and remove access from your identity provider groups; leavers lose access and ownership is flagged.
  • Access reviewsCross-hub reviews with independent reviewers, deadlines and sealed evidence.
  • Information barriersCompliance teams can build walls between groups of people or records.
  • Emergency access to sensitive fieldsA person who may see a record but not a sensitive field can request time-limited emergency access with a recorded reason.

Four-eyes and separation of duties

Sensitive changes need a second, different person.

7 features
  • Approval on sensitive actionsMoney, publication and policy changes route to approval and the maker cannot be the checker.
  • Separation-of-duties rulesDefine conflicting duties and block or flag them.
  • SoD waivers with expiryExceptions are time-limited and require a security approver.
  • Second approver for production configurationChange sets are approved by someone other than the submitter and can be gated before apply.
  • Risk-classified configurationSecurity-sensitive configuration kinds require step-up and a security approver.
  • Policy changes to AI need approvalWeakening an AI policy is treated as high risk.
  • Person-only decisionsAutomation and AI cannot approve, reject, delegate or release money, enforced in the database.

Sign-in and sessions

Strong identity with controls you can tune.

7 features
  • Single sign-on, SCIM and multi-factorStandards-based sign-in and provisioning.
  • Step-up re-authenticationHigh-risk actions need a fresh one-time code or passkey, and a client cannot assert its own step-up.
  • Session limits by roleStricter idle limits for administrators, with per-role overrides.
  • Sign-in history and anomaly findingsNew device, new country and impossible travel are flagged with both accesses side by side; the system never locks anyone out on its own.
  • Phone one-time codesVerified phone sign-in for external parties.
  • PasskeysPasskey support, switched off by default until enabled.
  • Throttling of sign-in attemptsRate limits on authentication.

Audit

A trustworthy record of who did what.

6 features
  • Append-only audit logEvery important write records actor, source, request and before and after; updates and deletes are refused for every role.
  • AI as tool, person as actorAudit distinguishes people, services and AI channels.
  • Append-only decision historyApprovals keep the rule version and a digest of the record.
  • Configuration auditEvery configuration write is attributed.
  • Audit export over the APIRead audit entries with changed field names only.
  • Tamper evidence by hash chainPlannedCryptographic chaining of audit entries.

Vendor access and AI controls

Even the vendor and AI work within limits.

9 features
  • Delegated support accessGrant a named engineer time-limited, scoped access with a purpose, a second approver by policy and an activity log.
  • Zero standing access designThe vendor works from metadata and artefacts, not customer business data.
  • AI kill switchStop one agent or all AI for a tenant, including calls in flight.
  • AI policy by role, field and regionAutonomy, field exclusions, providers and regions are administrator choices.
  • Agent registryUnregistered agents are refused on every entry point.
  • Secrets are write-onlyStored secrets are bound to a host and never shown again, with rotation and overlap.
  • Safe outbound requestsOne guard protects webhooks and connectors against server-side request forgery.
  • Signed webhooks and requestsReceivers can verify what  AI sends.
  • Operations consoleDead letters, webhook health and failed jobs in one place, with person-only replay.

Security centre and compliance programme

Continuous detection and proof.

5 features
  • Pre-publication scanning of customer codePlannedA gate that scans customer-authored scripts and configuration before publish.
  • Runtime detection rulesPlannedDetect abuse, exfiltration and anomalous agents from metadata events.
  • Containment actions and incident workflowPlannedOne-click and automatic containment with customer notification.
  • Customer security dashboard and log exportPlannedHealth score, alerts, findings and export to a security monitoring tool.
  • Fleet hygiene and compliance evidencePlannedVersion, certificate, backup and control evidence.
Assistant

What  AI prepares here

It prepares the work. A person confirms it.

  • Explain why a person has or lacks access
  • Prepare an access review
  • Prepare a role or sharing rule change for approval
  • Preview a uniqueness rule
  • Prepare a support access grant for a person to approve
  • Show AI activity and cost
Inside every hub

Workflow, reporting, integration, migration

The same four building blocks, specific to Security & Trust.

Workflow

Security-sensitive configuration, SoD waivers, support access and agent approvals route through the shared workflow with separation of duties.

  • SoD waiver needs a security approver
  • Support access above a window needs a second approver
  • Production change set needs a different approver
  • Higher-risk agent needs approval before running

Reporting and analytics

Access, sign-in and agent activity are available as governed datasets.

Datasets

Audit logAccess activityAgent activity

Dashboards

Access review progressAI activity

Integration

Audit entries, users and roles are readable over the API; sign-in and provisioning use standards.

/v1 audit entries, platform users and roles

Events

Access review events

Connectors

Single sign-onSCIM provisioning

Migration

Not applicable.

Mobile

On the phone

Remote wipe, device vault, certificate pinning and biometric re-lock protect the app; see the Mobile page.

Who uses it

Security administratorCompliance officerAuditorWorkspace ownerIT administratorSupport engineer (delegated)
Questions

Frequently asked

How is tenant data separated?

By row-level security enforced in the database, backed by automated isolation tests.

Can an administrator read everything?

No. Sensitive fields stay hidden until explicitly granted, and support access is time-limited, approved and logged.

Can AI see data a user cannot?

No. AI acts as the signed-in person with field exclusions you set, and hidden fields are never sent.

Can the audit log be edited?

Not by any role. Updates and deletes are refused.

Do you hold security certifications?

This page makes no certification claim. Ask for current attestations.

Can a maker approve their own change?

No. Approval and sensitive changes need a different person.

Can we erase a person's data?

Yes for sealed sensitive data, subject to legal holds and open disputes.

See Security & Trust on  AI

Talk to us