Quantum platform security should be engineered into the product, not added as an approval step after researchers already depend on informal workflows. Secure software development frameworks exist because most software development lifecycles need explicit security practices integrated into them [R244]. For quantum platforms, that integration must cover software, firmware, calibration artifacts, pulse libraries, runtime plugins, experiment data, and provider credentials.
View diagram source
flowchart LR
Idea[Platform idea] --> Threat[Threat model]
Threat --> Design[Secure design]
Design --> Build[Build and sign]
Build --> Test[Security and quantum validation]
Test --> Release[Progressive release]
Release --> Monitor[Continuous monitoring]
Monitor --> Improve[Root-cause improvement]
Improve --> ThreatDesign before approval
A secure-by-design quantum platform does not make the customer or operator compensate for unsafe defaults. The broker should reject unsafe combinations by default: unrestricted backend access, unpinned runtime dependencies, unreviewed plugins, missing evidence artifacts, ambiguous retention classes, or workload identities that cannot be traced to an accountable owner.
View diagram source
flowchart TB
Defaults[Secure defaults] --> Deny[Deny by default]
Defaults --> Least[Least privilege]
Defaults --> Pinned[Pinned runtime and target contracts]
Defaults --> Evidence[Evidence required]
Defaults --> Telemetry[Telemetry always on]
Defaults --> Rollback[Rollback path required]CISA’s secure-by-design guidance emphasizes that software manufacturers should make security a core design responsibility rather than shifting avoidable security burden to customers [R245]. In a quantum platform, the manufacturer may be the internal platform team. The “customer” may be a chemistry group, optimization team, financial-risk group, or external user of a managed quantum service.
Secure design review surfaces
| Surface | Design question |
|---|---|
| runtime | Can callers choose unsafe modes without policy approval? |
| compiler | Can optimization change observable semantics without equivalence evidence? |
| calibration | Can unreviewed calibration baselines be used for production results? |
| plugins | Can third-party code access credentials, raw results, or model artifacts? |
| data | Are retention, residency, and classification enforced before execution? |
| AI copilots | Can generated recommendations mutate production configuration? |
View diagram source
flowchart LR
Review[Secure design review] --> Runtime[Runtime surface]
Review --> Compiler[Compiler surface]
Review --> Calibration[Calibration surface]
Review --> Plugin[Plugin surface]
Review --> Data[Data surface]
Review --> Agent[Agent surface]Security requirements as product requirements
Security requirements should be written as product behavior, not generic control language.
| Requirement | Product behavior |
|---|---|
| traceability | every quantum result links to workload identity, commit, runtime, and backend |
| least privilege | users see only entitled providers, devices, queues, and evidence rooms |
| safe extensibility | plugins run in bounded sandboxes with explicit data grants |
| rollback | every runtime release has a tested downgrade or traffic-drain plan |
| PQC readiness | cryptographic inventory and migration status are tracked for platform dependencies |
CISA has identified support for standardized post-quantum cryptographic algorithms, consistent with NIST guidance, as an important product-security expectation [R245]. A quantum platform should therefore treat PQC readiness as part of its own supply-chain security posture, not merely as a customer-facing advisory.
View diagram source
journey
title Secure-by-design platform journey
section Before build
threat model: 5: Security, Platform
misuse cases: 5: Security, Product
section Build
signed artifacts: 5: Platform
dependency gates: 4: Platform
section Release
policy simulation: 5: Security
progressive rollout: 4: SRE
section Operate
continuous probes: 5: SRE
customer evidence: 4: ProductEngineering guardrails
Guardrails should be machine-checkable.
View diagram source
flowchart TB
PR[Pull request] --> Static[Static checks]
Static --> SBOM[SBOM and dependency review]
SBOM --> Policy[Policy simulation]
Policy --> Quantum[Quantum regression suite]
Quantum --> Evidence[Evidence schema validation]
Evidence --> Approval{Release allowed?}
Approval -- yes --> Release[Release train]
Approval -- no --> Fix[Fix or risk acceptance]Anti-patterns
View diagram source
mindmap
root((Insecure quantum platform))
Shared tokens
Notebook-only workflows
Unpinned SDKs
Unreviewed pulse changes
Missing provenance
Manual evidence assembly
Provider-specific exceptions
AI agents with write accessOperating rule
Security is not a gate at the edge of quantum operations. It is a property of the control loop. Every design choice should reduce the amount of manual trust required to run, verify, and explain a quantum result.