Oten Access

SECURITY & TRUST MODEL

Trust is proven, bounded, and recorded.

Oten Access does not grant implicit trust because a user is on a particular network or a device belongs to the organization. Decisions combine identity, device evidence, resource context, policy, and session state, then record whether enforcement was actually applied.

System role: Security architecture and product assurance

Only qualified evidence can produce bounded authority. The system records the decision, enforcement acknowledgements, and evidence required for safe recovery.

Security behavior is explicit at each trust boundary.

No network-location trust

IP address and location can be context; neither is sufficient authority to access a resource.

Two identities

User or workload identity and device identity are established and revoked independently.

Least privilege by resource

Policy scopes access to application, service, host, protocol, port, method, role, and time.

Bounded and revocable

Sessions, credentials, certificates, decisions, and connection leases have explicit lifetime and revoke semantics.

Signed desired state

Endpoint and Gateway validate immutable policy and configuration before activation.

Assume partial failure

Last-known-good, expiry, rollback, fail-open, fail-closed, and break-glass are explicit policy choices.

Audit decision and enforcement

The trail records policy version, target enforcement points, acknowledgements, failures, and recovery requirements, not only login.

Every identity and credential has a purpose, storage boundary, and revoke path.

User OIDC session

System-browser Authorization Code with PKCE, state and nonce validation, exact redirects, secure token storage, and separate local, IdP, and resource-session revoke semantics.

Device key or certificate

Binds an enrolled Agent Node; uses non-exportable hardware or platform storage only where support is verified; rotates or revokes independently from the user.

Setup Key

Bootstraps headless enrollment with one-time, TTL, usage, group, and audit constraints. It is not the long-term identity of the enrolled node.

WireGuard key

Authenticates encrypted data-plane peers under secure Agent storage and rotation policy; node removal updates authorized peer state.

Gateway node identity

Authenticates a data-plane node to Control Plane through mTLS or an equivalent mechanism, with rotation and explicit revoke.

Configuration signing key

Signs desired-state artifacts through a KMS or HSM boundary separated from the online serving role and supported by trust-bundle rotation.

JIT credential

A short-lived credential or certificate for scoped privileged authority with TTL, explicit revoke, and protocol-specific session termination.

Confidentiality does not replace configuration integrity.

  • Use immutable policy and configuration versions with canonical bytes, digest, and signature.
  • Bind artifacts to tenant, deployment, group, node audience, issue time, validity, and expiry.
  • Prevent rollback through a monotonic sequence or equivalent trusted mechanism.
  • Run schema, reference, compatibility, and security preflight before activation.
  • Expose Staged → Active transitions, health confirmation, bounded last-known-good, and recoverable rollback.
  • Audit actor, target, artifact version, node result, and rollback reason.

Continuous trust requires truthful evidence.

No false green

Process presence, test fixtures, mock data, stale signals, or an allocated interface cannot produce Verified, Protected, Connected, or Enforcing.

Evidence qualification

A trust signal identifies source, observation time, validity, policy generation, release, tenant and device binding, evidence digest, confidence, and reasons.

Mandatory gates

Revoked identity, invalid signature, expired authority, missing fail-closed guard, sensor quarantine, or critical verdict overrides a high Health Score.

Enforcement confirmation

The loop closes only when each required Policy Enforcement Point acknowledges the versioned action or reports failure.

Mode-specific revocation

Network flow, L7 request, WebSocket, SSH, database, Kubernetes, and PAM sessions have different revoke and termination semantics.

Evidence-based recovery

Restore requires new live evidence, resolved mandatory gates, hysteresis or approval, a new bounded decision, and confirmed enforcement.

Security evidence is governed by purpose and minimum necessary collection.

  • Collect only the posture, telemetry, metadata, and content evidence required by policy.
  • Separate product analytics from security evidence and never send identifiers, tokens, or user-entered sensitive text to analytics.
  • Enforce tenant isolation across storage, cache, queue, indexing, and search boundaries.
  • Use role-specific access to device detail, alert evidence, privileged-session recordings, and DLP incidents.
  • Make retention, export, deletion, redaction, and legal governance explicit.
  • Disclose Relay metadata and Gateway access-log scope without exposing protected infrastructure details.

Failure converges toward the configured restrictive state.

Critical resource + expired trust

Fail closed with a safe reason, last synchronization, expiry, request ID, and correlated audit.

Low-risk resource + short outage

Policy may permit bounded last-known-good behavior with a visible offline or grace state and expiry.

Unknown posture provider

Unknown or stale remains explicit and produces the configured policy outcome rather than an implicit pass.

Invalid configuration signature

Reject activation, preserve a valid last-known-good artifact or restrictive baseline, stop rollout, and raise a high-severity event.

Audit destination unavailable

Buffer within a defined limit, surface backpressure, and never silently lose evidence indefinitely.

Break-glass

Use explicit time-bound authority, strong authentication, immediate notification, enhanced audit, and post-event review.

Guardrails address identity, policy, data path, and operational threats.

Enrollment abuse

One-time or bounded Setup Keys, approval, organization binding, hardware evidence where supported, and operator separation.

Token or credential theft

PKCE, secure storage, rotation, revoke, device policy, short lifetime, and audience validation.

Configuration tampering or replay

Signature, audience binding, expiry, trusted sequence, preflight, and bounded last-known-good.

Signal or Relay compromise

Authenticated channels, replay protection, peer authentication, end-to-end encryption, metadata minimization, rate limits, and audit.

Gateway compromise or split brain

Node identity, least privilege, secret isolation, explicit Failover Set, compatibility, fencing, rotation, and quarantine.

Overbroad policy or telemetry poisoning

Default deny, resource scope, impact preview, dry run, source identity, schema validation, confidence, reasons, and response guardrails.

Standards alignment is not certification.

NIST SP 800-207

The architecture follows identity- and resource-focused Zero Trust principles and avoids network-location implicit trust.

OpenID Connect

Provides the identity layer for federation and native-client authentication.

RFC 7636 / PKCE

Reduces authorization-code interception risk for public native clients as one part of the full OIDC security profile.

WireGuard

Provides the encrypted overlay data plane; resource authorization remains an independent identity and policy decision.

FIPS 203 / ML-KEM

Provides a standardized primitive for hybrid cryptography design. FIPS validation and quantum-resistance claims remain separate assurance properties with their own scoped evidence.

Build confidence from evidence, not implied trust.

Review the technical dependencies and evidence requirements for active revocation, platform assurance, Gateway HA, and hybrid cryptography.