Join our Newsletter — 33% off our NHI Course
Home› Guides› OWASP SAMM Guide: Assess and Raise Software Security…
Guide Application Security

OWASP SAMM Guide: Assess and Raise Software Security Maturity

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 27 October 2026 14 min read
On this page

Most software security programmes cannot say, with evidence, how mature they are or what to improve next. OWASP SAMM (Software Assurance Maturity Model) is the OWASP flagship model for answering that question: it breaks secure software work into 15 practices, scores each on three maturity levels and turns the gaps into a roadmap. This guide goes deeper than our general AppSec maturity overview: it explains the model structure stream by stream, how the scoring really works, how to run an assessment that produces defensible scores, and how to use SAMM alongside NIST SSDF, ISO/IEC 27001 and SP 800-53. It is written for AppSec leads, security architects and engineering managers who own a software security programme.

Key takeaways

  • Five functions, 15 practices, 30 streams. SAMM v2 groups activities under Governance, Design, Implementation, Verification and Operations; every practice has two streams (A and B) and each stream has three maturity levels.
  • Scores are about coverage and quality. Each activity has one assessment question with quality criteria; if the criteria are not completely met the answer is "No", and partial answers record how much of the application estate is covered.
  • Scope decides whether the score means anything. Assess a defined unit (a product line, a business unit or the whole organisation) and keep the same scope for reassessment, or trend lines are meaningless.
  • Targets beat maximum scores. SAMM is designed for a risk-based target per practice; level 3 everywhere is rarely a sensible goal for a small or low-risk team.
  • Machine identities sit inside the model. Secret management (Secure Deployment stream B), build integrity and dependency review are where non-human identity controls show up in SAMM, and where many software supply chain incidents begin.

What SAMM is and how it differs from other maturity models

SAMM is an open, vendor-neutral, prescriptive model maintained by the OWASP SAMM project. The website lists version 2.0 as the current model (checked 27 October 2026); the raw model is published as YAML in the project's core repository on GitHub, from which the website, PDF and toolbox are generated. Version 2 restructured the older 1.x model into five business functions and added the Implementation function, which covers build, deployment and defect management.

It is prescriptive in the sense that every level of every stream names a concrete activity ("Define a consistent build process", "Protect application secrets in configuration and code") with quality criteria an assessor can check. That is the main difference from a descriptive model such as BSIMM, which reports what observed firms actually do rather than prescribing what you should do. See the AppSec Maturity Model for how SAMM compares with other models and for a five-level maturity view of AppSec as a whole.

  • Use SAMM when: you want a self-assessment you can run internally, an improvement roadmap tied to named activities, and a model that covers development and operations, not just testing.
  • Use a descriptive benchmark when: leadership mainly wants to know how the organisation compares with peers. SAMM has its own Benchmark initiative, which collects shared assessment data so organisations can compare once enough data exists.
  • Use NIST SSDF or a regulatory standard when: you need to attest to a fixed set of practices (for example for US federal software suppliers). SAMM then becomes the engine for measuring how well you meet them.

Model structure: functions, practices and streams

The table lists every practice and its two streams as named in the SAMM core model. Stream A and stream B usually split a practice into "doing the activity" and "managing its inputs or feedback".

Business functionSecurity practiceStream AStream B
GovernanceStrategy and Metrics (SM)Create and PromoteMeasure and Improve
GovernancePolicy and Compliance (PC)Policy and StandardsCompliance Management
GovernanceEducation and Guidance (EG)Training and AwarenessOrganization and Culture
DesignThreat Assessment (TA)Application Risk ProfileThreat Modeling
DesignSecurity Requirements (SR)Software RequirementsSupplier Security
DesignSecure Architecture (SA)Architecture DesignTechnology Management
ImplementationSecure Build (SB)Build ProcessSoftware Dependencies
ImplementationSecure Deployment (SD)Deployment ProcessSecret Management
ImplementationDefect Management (DM)Defect TrackingMetrics and Feedback
VerificationArchitecture Assessment (AA)Architecture ValidationArchitecture Mitigation
VerificationRequirements-driven Testing (RT)Control VerificationMisuse/Abuse Testing
VerificationSecurity Testing (ST)Scalable BaselineDeep Understanding
OperationsIncident Management (IM)Incident DetectionIncident Response
OperationsEnvironment Management (EM)Configuration HardeningPatching and Updating
OperationsOperational Management (OM)Data ProtectionSystem Decommissioning / Legacy Management

Activities are identified by function, practice, level and stream, for example I-SB-1-A (Implementation, Secure Build, level 1, stream A). Using these identifiers in your roadmap and tickets makes reassessment much easier, because evidence can be filed against the exact activity it supports.

The three maturity levels in practice

The core model defines the levels in one sentence each: level 1 is an ad hoc, best effort implementation of an activity; level 2 is a consistent, repeatable process that can be relied on; level 3 maximises effectiveness through continuous improvement based on effective and timely feedback. The progression is easiest to see in a single stream.

StreamLevel 1Level 2Level 3
I-SB-A Build ProcessDefine a consistent build processAutomate the build processEnforce a security baseline during build
I-SB-B Software DependenciesIdentify application dependenciesReview application dependencies for securityTest application dependencies
I-SD-B Secret ManagementProtect application secrets in configuration and codeInclude application secrets during deploymentEnforce lifecycle management of application secrets
D-TA-B Threat ModelingPerform basic threat modelingStandardize and scale threat modelingOptimize threat modeling
V-ST-A Scalable BaselinePerform automated security testingDevelop application-specific security test casesIntegrate security testing tools in the delivery pipeline
I-DM-A Defect TrackingTrack security defects centrallyRate and track security defectsEnforce an SLA for defect management
G-SM-B Measure and ImproveDefine basic security metricsSet strategic KPIsDrive the security program through metrics

Two practical points follow. First, levels are cumulative within a stream: claiming level 3 for dependency testing while dependencies are not inventoried (level 1) produces a score nobody should trust. Second, the two streams of a practice can sit at different levels, which is normal and tells you more than a single practice number.

How SAMM scoring works

The core model contains 90 assessment questions, one per activity (15 practices, two streams, three levels). Each question carries quality criteria. For Secure Deployment stream B level 1, for example, the question asks whether access to application secrets is limited according to least privilege, and the criteria include storing production secrets in a secured location, developers having no access to production secrets, and production secrets not being available in non-production environments.

  • The "No" rule: the SAMM assessment page states that if the quality criteria are not completely fulfilled, the answer is "No". This single rule prevents most score inflation.
  • Coverage answers: when criteria are met, the answer records how widely. One standard answer set in the model uses "Yes, for some applications" (0.25), "Yes, for at least half of the applications" (0.5) and "Yes, for most or all of the applications" (1). Other answer sets use different wording for activities that are not per-application.
  • Roll-up: the toolbox Scorecard tab updates automatically from the Interview tab answers, giving a score per stream, practice and business function on the 0 to 3 scale.
  • Tooling: OWASP publishes the toolbox as an Excel and a Google Sheets spreadsheet, and SAMMwise as a self-hosted online tool on GitHub. A third-party online tool from a project sponsor also exists; the model and scores are the same whichever you use.

Because partial coverage earns fractional credit, an organisation with 200 applications can show real progress long before every team reaches a level. Report the coverage figure alongside the score so leadership can see whether a 1.5 means "half the estate at level 2" or "everything at level 1 and a few teams beyond".

Running an assessment that produces defensible scores

  1. Fix the scope. Decide whether you are scoring the organisation, a business unit or one product line, and write it down with the list of applications included. Mixed scopes are the most common reason two assessments cannot be compared.
  2. Read the guidance first. The project points assessors to its Quick Start Guide and Assessment Guide before starting. Agree internally how you will interpret the answer sets (what counts as "an application", for example).
  3. Choose interviewees per function. Governance questions go to the security and compliance leads; Design and Verification to architects, AppSec and QA; Implementation to platform and DevOps engineers; Operations to SRE, infrastructure and incident response. A single interviewee rarely knows all 90 answers.
  4. Collect evidence while you interview. For each "Yes", record the artefact that supports it: a pipeline definition, a secrets manager policy, a threat model, a defect SLA report. Store it against the activity ID.
  5. Calibrate. Have a second assessor review a sample of answers, especially every level 3 claim. Downgrade where evidence does not meet all criteria.
  6. Set targets per practice. Use the application risk profile (D-TA-A) and the organisation's risk appetite (G-SM-1-A) to set a target level per stream. High-risk, internet-facing products justify higher targets than internal tools.
  7. Build the roadmap in phases. The toolbox supports roadmap planning. Group improvements into phases of three to six months, each with a small number of streams moving one level.
  8. Reassess on a fixed cadence. Annual is typical for a full assessment; reassess individual streams after each roadmap phase. Optionally share anonymised results with the SAMM Benchmark initiative.

Rollback and correction: if a reassessment shows a stream has regressed (for example after a reorganisation removes security champions), record the regression honestly rather than carrying forward the old score. A model that only ever goes up is not being used as a measurement tool.

Decision guidance: targets for small and large organisations

SituationSensible approachTrade-off
Startup or single product teamAim for level 1 across all streams first, then level 2 in Secure Build, Secure Deployment and Security TestingGovernance streams stay light; acceptable while the team is small enough to share context
Mid-size firm, 20 to 100 applicationsOrganisation-wide scope, targets set per stream from risk profiles, roadmap phases of about six monthsCoverage answers will hold scores down until tooling is standardised
Large enterprise, many business unitsAssess per business unit with a common method, then aggregate; central team owns Governance practicesHigher assessment cost; calibration across assessors becomes essential
Regulated or federal supplierMap SAMM activities to the required framework (SSDF, PCI Secure Software Standard) and set level 2 or above where the framework requires repeatable practiceSome SAMM streams may matter less to the regulator; keep them for internal value
Outsourced developmentPrioritise Supplier Security (D-SR-B) and Requirements-driven TestingYou can only score what you can evidence from the supplier

Where identity and secrets fit in SAMM

SAMM is not an identity model, but several streams are where machine identity controls are measured, and they are worth scoring strictly because CI/CD credentials are a recurring entry point.

  • Secret Management (I-SD-B): level 1 protects secrets in configuration and code, level 2 injects secrets during deployment rather than storing them in artefacts, level 3 enforces lifecycle management (rotation, revocation, ownership). See the Secrets Management Guide.
  • Build Process (I-SB-A): the level 1 activity already requires that the build definition contains no secrets, that build tools are patched and hardened, and that artefacts carry an integrity value such as a hash or signature, with the signing key protected.
  • Software Dependencies (I-SB-B): inventory and review of third-party components, which is where an SBOM programme is measured. See the SBOM Guide.
  • Deployment Process (I-SD-A): level 3 verifies the integrity of deployment artefacts, which depends on protecting the service accounts and tokens that pipelines use to deploy.

When scoring these streams, ask who owns each pipeline credential, how it is rotated and what happens when an engineer leaves. Without clear answers, the criteria are not completely met and the answer is "No".

Mapping SAMM to other frameworks

SAMM is often used as the measuring tool for obligations defined elsewhere. The mapping below is indicative, based on the scope of each control, not an official crosswalk.

SAMM practiceNIST SP 800-53 Rev. 5NIST CSF 2.0ISO/IEC 27001:2022 Annex A
Strategy and Metrics, Policy and ComplianceSA-3 System Development Life CyclePR.PS-06A.5.1 Policies for information security
Threat Assessment, Secure ArchitectureRA-3 Risk Assessment, SA-8 Security and Privacy Engineering PrinciplesID.RA-01A.8.27 Secure system architecture and engineering principles
Security Requirements (Supplier Security)SR-3 Supply Chain Controls and ProcessesGV.SC-05A.5.20 Addressing information security within supplier agreements
Secure Build, Secure DeploymentSA-15 Development Process, Standards, and Tools; IA-5 Authenticator ManagementPR.PS-06A.8.25 Secure development life cycle
Security Testing, Requirements-driven TestingSA-11 Developer Testing and EvaluationPR.PS-06A.8.29 Security testing in development and acceptance
Defect Management, Environment ManagementSI-2 Flaw Remediation, CM-6 Configuration SettingsID.RA-01A.8.8 Management of technical vulnerabilities
Operational Management (decommissioning)SA-22 Unsupported System ComponentsPR.PS-06A.8.10 Information deletion

For the US federal software attestation context, see the NIST Secure Software Development Framework (SSDF, SP 800-218) Guide. In CIS Controls v8.1 the closest control is Control 16, Application Software Security.

Detection and measurement: proving the programme works

SAMM scores are lagging indicators; pair them with operational metrics that the Measure and Improve stream (G-SM-B) and Defect Management metrics stream (I-DM-B) expect you to define.

  • Score against target: number of streams at or above their target level, by business unit.
  • Coverage: percentage of in-scope applications covered by each level 1 activity (for example, percentage with a dependency inventory or automated security testing in the pipeline).
  • Defect SLA compliance: percentage of security defects fixed within the agreed SLA by severity, the evidence behind I-DM-3-A.
  • Secrets hygiene: secrets detected in source or build logs per month, and median time to revoke them.
  • Threat model currency: percentage of high-risk applications with a threat model reviewed in the last 12 months.
  • Roadmap delivery: planned stream improvements completed per phase.

Common failure patterns

  • Self-scoring without evidence. Teams answer "Yes" because a tool is licensed, not because criteria are met. The "No" rule exists to stop this; enforce it with evidence links.
  • Changing scope between assessments. Adding or removing business units changes coverage answers and makes improvement look like regression, or the reverse.
  • Targeting level 3 everywhere. This spreads effort thinly and ignores the risk-based targeting the model is built around.
  • Treating Secret Management as a tooling question. Buying a vault does not satisfy lifecycle management if pipeline tokens are never rotated. The Codecov Breach showed how a compromised CI script could collect the credentials held in customers' build environments, which is exactly the exposure the Build Process and Secret Management streams are meant to reduce.
  • Ignoring Operations. Development-led programmes often skip Environment Management and Operational Management, so decommissioned applications keep live credentials and unpatched hosts.
  • Assessment as a one-off project. Without a reassessment date and a roadmap owner, the scorecard becomes a slide that ages quickly.

Practitioner checklist

  • Write down the assessment scope and the list of in-scope applications before the first interview.
  • Download the current model and toolbox from the OWASP SAMM project and agree how the answer sets will be interpreted.
  • Interview at least one person per business function and record evidence against each activity ID.
  • Apply the "No" rule strictly and have a second assessor calibrate every level 3 answer.
  • Set a target level per stream from application risk profiles and risk appetite, not a blanket target.
  • Score Secret Management, Build Process and Software Dependencies with specific questions about pipeline credentials and signing keys.
  • Map practices to the frameworks you must evidence (SSDF, SP 800-53, ISO/IEC 27001) so one assessment serves several audits.
  • Publish a phased roadmap with owners and a reassessment date.
  • Track three to six operational metrics alongside the SAMM scores.

Standards and references

Related NHI Mgmt Group resources: AppSec Maturity Model · Application Security Guide · DevSecOps Guide · Security Champions Programme Guide · Threat Modelling Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 27 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.