Security Obligations

Providers hold two kinds of trust: tenants trust you with their workloads, and the protocol trusts your signatures. Both are revocable. This page collects the security obligations spread across the provider tooling.

Key management

• The provider signing key is your identity. Every usage record, settlement transaction, and node-metadata update is signed with it. Ship it via PROVIDER_SIGNING_KEY_PATH from a secret store — never in config files or images.
• Node agent keys are per-node. Each hpc-node-agent has its own Ed25519 key pair; register public keys explicitly so a compromised node can be individually deregistered.
• The repository’s DR tooling includes key backup (scripts/dr/backup-keys.sh --type all) — use it, and store backups offline.

Credential hygiene

All scheduler and platform credentials come from the environment:

export SLURM_JWT_TOKEN="..."
export MOAB_USERNAME="..."
export MOAB_PASSWORD="..."
export OOD_OIDC_CLIENT_SECRET="..."

In Kubernetes deployments, secrets resolve from external secret stores (secret/virtengine/{env}/{service} paths via ESO/Vault in the reference topology) — pods must not carry inline secrets.

For MOAB over SSH, pin host keys: ssh_host_key_callback: known_hosts (or pinned); never ship insecure to production.

Workload isolation

• Enforce template security policy: registry allowlists, required image digests, sandbox_level: strict, and host-mount allowlists (see HPC Workload Templates).
• Keep tenant workloads off validator infrastructure. The reference deployment separates them with node labels and taints (role=validator + dedicated=validator:NoSchedule vs. role=workload).
• Tenant↔provider traffic is authenticated with x/cert mTLS certificates.

Audit logging

Enable the daemon’s audit log and keep it:

audit:
  enabled: true
  log_path: /var/log/virtengine/hpc-audit.log
  log_job_events: true
  log_security_events: true
  log_usage_reports: true

Audit records are your evidence in disputes and fraud reviews — the protocol records its own view on-chain, and you want a matching local trail.

Routing and enforcement posture

Run routing enforcement in strict mode in production. permissive and audit_only exist for migration and monitoring windows, not steady state — unauthorized fallbacks are a recorded violation type for a reason.

What the protocol does if you don’t

Failure	Protocol response
Bad or unsigned usage records	Rejected before settlement
Reconciliation drift > 25%	Settlement blocked
Anomalous usage patterns	Fraud reports via x/fraud; lease enforcement
Audit lapses	Loss of audit-gated order eligibility
Sustained disputes	Reputation damage via x/review, dispute history