For the complete documentation index, see llms.txt. This page is also available as Markdown.

Security architecture

Security model

Decentralization Manager combines host controls, operator authentication, Canton authorization, encrypted peer communication, protected local state, supply-chain controls, and on-ledger governance. No single layer secures the whole deployment.

Each organization operates its own instance and controls its accounts, identity provider, Canton credentials, local storage, network exposure, peer records, and upgrades. On-ledger governance limits application actions; it does not protect a compromised host or incorrect off-ledger input.

Trust boundaries

Treat every arrow as a separate boundary requiring the appropriate authentication, authorization, encryption, validation, logging, and failure handling.

Control ownership

Boundary
Primary controls
Canonical procedure

Operator access

OIDC, PKCE, role enforcement, session and account controls

Canton connections

Separate Admin and Ledger credentials, least privilege, restricted network paths, TLS or mTLS

Peer communication

Persistent Noise identity, encrypted transport, allowlisted keys, out-of-band verification

Persistent data

Restricted storage, database encryption, protected backups, consistent recovery

DAR supply chain

Approved source, version pinning, hashes, review, and package comparison

Governance

On-ledger proposals, confirmations, active rules, execution, expiry, and result verification

Release compatibility

Approved versions, peer protocol, migrations, known issues, and support status

Threat model and residual risks

  • Compromised operator account: can invoke the routes and Canton rights available to that identity.

  • Compromised host: can expose local credentials, falsify the UI, interrupt workflows, or misuse assigned Canton rights.

  • Wrong peer record: creates encrypted trust with the wrong endpoint; encryption cannot replace identity verification.

  • Lost or replaced Noise identity: changes how peers recognize the instance and requires coordinated replacement.

  • Malicious or incompatible DAR: can affect application behavior despite successful distribution.

  • Mixed peer versions: can interrupt coordination when protocol compatibility changes.

  • Coordinator failure: can stop workflow progress without invalidating the existing Decentralized Party.

  • Inconsistent restore: can preserve only part of the identity, credential, package, or workflow state.

  • Sensitive logs or support exports: can leak credentials or confidential identifiers if not sanitized.

Governance limits

Governance authorizes application actions on Canton; it does not prove that every input, package, host, or external system is safe. Members must review the proposal, referenced contracts, package versions, and expected effects before confirming.

Do not use a fixed approval count as a universal product property. Production rules and any intentional design constraints must be documented for the supported release.

Audit and disclosure

Audit pages identify the reviewed commit, components, dates, exclusions, design acknowledgements, and remediation status. An audit is point-in-time evidence, not a guarantee for every deployment or future release.

Use one approved private vulnerability-reporting route. Do not report unpatched vulnerabilities through public issues or discussions.

Technical sources

Next step

👉 Apply the controls: Decentralization Manager production deployment.

Last updated