SECURITY + TRUST

Make trust visible.

Commonset is designed to govern AI capabilities that may contain valuable organizational logic, instructions, files, and integrations. Security controls should be understandable, attributable, and enforceable—not hidden behind a generic trust badge.

Our security model focuses on four things: tenant isolation, capability confidentiality and integrity, credential confidentiality, and governance integrity across every interface and provider adapter.

01

Tenant isolation

A principal can act only inside an organization where Commonset currently authorizes that user, session, token, or OAuth grant.

02

Capability integrity

Capability content, versions, metadata, findings, and approval state stay bound to the authorized organization and recorded artifact.

03

Credential confidentiality

Provider secrets, signing material, sessions, bearer tokens, and encryption keys are treated as credentials—not ordinary metadata.

04

Governance integrity

Review, approval, separation of duties, access, provenance, and trust decisions should not be bypassable through a provider-specific path.

CAPABILITY PROTECTION

Protect the capability itself.

Commonset treats capability content as a governed asset. Confidentiality and integrity controls sit above the underlying storage provider so the organizational security model does not depend on one infrastructure vendor.

CONTENT

Encrypted before storage

Trusted capability files are encrypted at the application layer using organization-specific key material before bytes reach the configured storage backend.

METADATA

Tenant-bound sensitive fields

Selected sensitive metadata uses authenticated encryption bound to the organization and record context rather than relying only on database-at-rest encryption.

CREDENTIALS

Separate credential boundary

Connected-provider credentials are stored as encrypted ciphertext and decrypted only inside authorized provider operations.

INTEGRITY

Recorded content digests

Immutable versions retain content digests and provenance so Commonset can verify that stored artifact bytes still match the registered version.

UNTRUSTED BY DEFAULT

Uploaded capability packages earn trust before promotion.

ZIP uploads begin in quarantine. Commonset validates the archive structure and capability format, applies malware policy and static security analysis, and records evidence before content moves into trusted storage.

The registry does not execute uploaded scripts. Static analysis is evidence for governance and human review; it is not a sandbox and does not prove that a capability is safe.

01QuarantineOwner-only temporary storage
02Archive controlsTraversal, symlinks, nested archives, size, compression
03Security analysisMalware, secret-like material, executable and behavioral signals
04Trusted versionValidated content, digest, findings, provenance

IDENTITY + AUTHORIZATION

Authentication proves identity. Commonset still decides access.

External identity providers do not automatically grant tenant access. Commonset keeps organization membership, capability permissions, grants, and lifecycle policy in its own authorization model.

Browser and organization access

  • Organization context is derived from current Commonset membership.
  • Capability and administrative views remain scoped to the active organization.
  • Capability permissions distinguish discover, use, review, and manage.
  • External identity is resolved through current provider connection, Commonset user, membership, and organization state.

MCP and connected AI clients

  • Developer bearer tokens are stored as hashes and revalidated against current membership, expiry, and revocation.
  • MCP OAuth access is checked for issuer, resource audience, external organization, concrete client, active Commonset grant, and allowed scopes.
  • MCP tools apply the same Commonset capability permissions and trust-availability rules used by browser workflows.
  • Revoking a Commonset OAuth grant removes that client's Commonset authorization path.

GOVERNANCE + AUDITABILITY

Trust is a set of decisions with evidence behind them.

Commonset separates technical findings from organizational approval. A capability can be structurally valid and still require review before it becomes available to a wider audience.

Immutable versions

Changes create versioned artifacts rather than silently replacing an approved capability.

Provenance

Source, parent version, actor, origin metadata, and artifact digest are recorded with the version.

Separation of duties

Organizations can require a reviewer other than the creator before broader approval.

Policy gates

Blocked, unresolved, or stale-trust versions can be prevented from approval, access, download, or provider publication.

Material audit events

Creation, review, approval, access, provider activity, MCP activity, and administrative changes generate attributable events where implemented.

Provider-neutral control

A provider integration adapts the capability; it does not replace Commonset membership, grants, review state, or policy.

APPLICATION + RELEASE SAFEGUARDS

Security also depends on predictable operation.

Commonset's deployment path is being hardened around explicit configuration, deterministic checks, visible release identity, and repeatable smoke verification.

WEB

Browser protectionsDjango CSRF protection for state-changing browser workflows, a restrictive content security policy, and no-store caching for authenticated and sensitive paths.

CONFIG

Fail-closed Production contractProduction startup rejects unsafe development settings, missing encryption material, insecure URLs, local-admin shortcuts, and other unsafe overrides.

RELEASE

Traceable deploymentsThe live health endpoint exposes environment and exact release SHA so operators can identify what code is serving traffic.

VERIFY

Repeatable release smoke checksA release-smoke command validates deployment checks, database connectivity, migrations, storage connectivity, malware posture, email configuration, and release identity without creating customer data.

CURRENT MATURITY

Precise about what exists today.

Commonset is currently an early-stage product preparing for design-partner and pilot use. We would rather describe current controls accurately than imply a certification or operating maturity we have not yet earned.

CURRENT

Implemented foundations

  • Tenant-aware application authorization
  • Encrypted capability content and sensitive metadata paths
  • Encrypted provider credentials
  • Untrusted upload and scanner controls
  • Versioned governance, access, provenance, and audit foundations
  • MCP bearer-token and OAuth authorization controls
  • Explicit Production configuration and release checks
NOT CLAIMED YET

Enterprise validation still to complete

  • Commonset does not currently claim SOC 2 certification.
  • An independent penetration test has not yet been completed.
  • Production infrastructure, backup/restore validation, and intermediary logging controls will be verified before real customer Production traffic.
  • SCIM, SIEM integrations, customer-managed encryption/storage, and formal enterprise compliance programs are future enterprise work.

SECURITY EVALUATION

Bring your requirements.

If you're evaluating Commonset as a design partner, we can walk through the current architecture, trust boundaries, controls, and gaps directly—without hiding behind a generic security badge.

Discuss security with Commonset