Skip to content
SecHelixv3 alpha
GitHub
DocsContributeSupportWorkbenchGitHub
Teams

Enterprise Adoption

A four-phase rollout: baseline one service, add a private policy pack, wire the gate into CI and pull requests, then measure the metrics that matter.

Phase 1 — baseline one service

SecHelix is designed to coexist with the SAST, SCA, DAST, browser, runtime, cloud, and in-house tooling you already run. It coordinates their evidence under one verification standard rather than replacing them.

Choose one representative repository and run in STATIC, then LOCAL if a safe runtime is available. Capture:

  • The attack surface and trust boundaries.
  • The role by object by action matrix.
  • Applicable, unknown, and blocked coverage.
  • Verified findings and rejected candidates.
  • Security regression tests.
  • The release-gate outcome.
  • Operator time and manual blockers.

Phase 2 — organization policy pack

Add private requirements without forking the core evidence contract:

  • Mandatory authentication and session controls.
  • Tenant-isolation invariants.
  • Company-specific sensitive-data rules.
  • Payment and accounting invariants.
  • Approved and forbidden dependencies or providers.
  • Cloud and IAM expectations.
  • Required log and audit events.
  • Production-safe test restrictions.
  • Escalation and risk-acceptance owners.

Phase 3 — CI and pull requests

Use SecHelix as an evidence-aware review layer around existing CI. The recommended policy:

  1. Block on verified regressionsVerified Critical and High security regressions block the release.
  2. Fail closedMalformed or missing required evidence fails closed to INCOMPLETE.
  3. Keep uncertainty visibleUNKNOWN and BLOCKED remain visible and cannot silently become NOT_APPLICABLE.
  4. Own accepted riskAccepted risk requires an owner, a reason, and an expiry or review date.
  5. Require regression proofSecurity fixes carry regression proof where practical.

Phase 4 — measurement

Track operational metrics rather than raw finding counts.

MetricWhat it tells you
Verified precisionWhether reported findings survive scrutiny.
False-positive rejection rateWhether verification is doing work.
Detection recall on known fixturesWhether the workflow finds what it should.
Time to verificationHow long a candidate stays ambiguous.
Time to root-cause fixWhether repairs land at the boundary or the symptom.
Regression-proof rateWhether fixes are provably closed.
Surface classified vs unknown or blockedHow much of the system you can currently assess.
Recurrence rate of fixed classesWhether canonical fixes actually hold.

Roles

A mature deployment separates these responsibilities:

Mapper
Architecture and trust boundaries.
Specialists
AuthZ, authN, business logic, injection, supply chain, cloud, AI and MCP.
Runtime verifier
Evidence from browser, API, database, and test environments.
Independent verifier
Tries to disprove important candidates.
Owner / release authority
Accepts risk and makes release decisions.

Do not let model reputation replace independent evidence.

Procurement questions

The seven questions the framework recommends a security review ask, with its own answers:

QuestionAnswer
Can we inspect and version the full methodology?Yes. The core is open source under Apache-2.0.
Are unsupported scanner or model claims reported as confirmed vulnerabilities?They should not be. That is the central contract.
Can we retain private policy and evidence?Organization-specific policy can remain private.
Can it coexist with existing scanners?Yes. Adapters normalize evidence rather than requiring replacement.
Does it claim measured accuracy?One blind label-only run is published (precision 0.950). Full-workflow accuracy is NOT_MEASURED.
Can it make fail-closed release decisions?Yes: PASS, PASS_WITH_KNOWN_RISK, BLOCKED, and INCOMPLETE are defined and enforced.
Is third-party active testing automatically authorized?No. Scope and authorization are explicit prerequisites.

A pilot is useful if it catches a real boundary failure existing review missed, rejects a meaningful false positive that existing automation escalated, finds sibling variants after one root cause, converts a recurring class into a regression test or policy, or produces a clearer release decision than disconnected scanner outputs.