V-Standard 2026.3

Supported protocol versions and cipher suites, certificate chain, expiry, revocation information, key strength and known TLS weaknesses on every public endpoint in scope.

Plain-HTTP requests are redirected to HTTPS, HSTS is present with an adequate lifetime, and no content in scope is served over an unencrypted channel.

Authoritative servers, DNSSEC where applicable, dangling records that allow sub-domain takeover, zone transfer exposure and consistency between public and expected records.

Sender authentication for the product's domains: SPF syntax and scope, DKIM signing, DMARC policy and reporting, MTA-STS where present.

Content-Security-Policy, frame protection, content-type sniffing, referrer policy, permissions policy and cache directives on sensitive responses.

Inventory of listening services on the public perimeter, unexpected management ports, databases or message brokers reachable from the internet.

Identified server software, frameworks and components on the perimeter are matched against published vulnerability records (CVE/NVD, CISA KEV, vendor advisories).

Administration panels, monitoring dashboards and management consoles reachable from the internet; default or vendor credentials on any such interface.

Version banners, directory listings, debug endpoints, source-control folders, environment files, backups and stack traces reachable on the perimeter.

Secure, HttpOnly and SameSite attributes on session and sensitive cookies, cookie scope and lifetime.

Open redirects, insecure downgrade paths and mixed-content resources on pages in scope.

A published way to report a vulnerability (security.txt or equivalent) that reaches a responsible person and is answered.

Credential policy, multi-factor availability, brute-force and credential-stuffing protection, safe handling of failed logins and enumeration.

Session identifier generation, rotation on privilege change, expiry, revocation on logout and concurrent-session handling.

Role model enforcement on every function and object: horizontal and vertical privilege escalation, insecure direct object references, mass assignment.

Separation of data between organisations or accounts in a multi-tenant product: queries, storage, caches, background jobs and exports.

Server-side injection classes — SQL, NoSQL, command, template, path traversal, server-side request forgery, XML external entities — across the endpoints in scope.

Cross-site scripting (reflected, stored, DOM-based), cross-site request forgery, clickjacking and unsafe client-side storage.

Token handling, key rotation, object- and property-level authorisation on API resources, function-level access control.

Undocumented or deprecated endpoints, debug and internal APIs reachable from outside, excessive data exposure in responses.

Limits on authentication, messaging, export and other resource-intensive operations; protection against automated abuse.

Type validation, size limits, storage location, execution prevention and safe processing of uploaded content.

Error responses do not leak internals; logs do not contain secrets or excessive personal data; failures are recorded.

Workflow bypass, price or quantity manipulation, replay of one-time actions and race conditions in the key flows in scope.

Password reset, email change and account recovery: token strength, expiry, single use and protection against takeover.

Webhook signature verification, OAuth flows, callback URL validation and least-privilege scopes for integrated services.

Software composition analysis of direct and transitive dependencies against published vulnerability records; unmaintained and end-of-life components.

Credentials, tokens, keys and certificates in the repository tree, commit history, build configuration and container layers.

Rule-based analysis of the source for injection, unsafe deserialisation, cryptographic misuse, insecure defaults and dangerous functions, with manual triage of findings.

A machine-readable SBOM (CycloneDX or SPDX) exists, matches the deployed build and can be regenerated by the pipeline.

Pipeline permissions, secret handling in jobs, third-party actions and images, artefact signing and separation between build and deploy credentials.

Lock files, version pinning, integrity hashes and controlled registries for packages, base images and build tools.

Base image currency, vulnerable packages, secrets in layers, running as root and unnecessary capabilities.

Protected default branches, mandatory review, signed commits or verified authorship, and restricted force-push.

Reproducible or attested builds, artefact signatures and verification before deployment.

Root and administrative accounts, MFA enforcement, role scoping, long-lived keys and cross-account trust in the cloud accounts in scope.

Object storage, snapshots, databases and file shares reachable without authentication or with over-permissive policies.

Security groups, network ACLs, private subnets for data stores, bastion access and east-west restrictions.

Central secret storage, rotation, access auditing and absence of secrets in configuration files, images or environment dumps.

Defined patch cadence, tracking of published vulnerabilities for the stack in use and evidence that Critical fixes are applied within the declared window.

Security-relevant events are logged centrally with retention, alerting on authentication anomalies and administrative actions, and protected from tampering.

Backup coverage of production data, encryption of backups, off-site copies, restoration tests and recovery time objectives.

Operating system and service configuration against a documented baseline, unnecessary services disabled, administrative access restricted.

Encryption of databases, object storage, backups and disks holding customer data; algorithms and key ownership.

Encryption of internal service-to-service traffic, database connections and administrative channels, not only the public edge.

Generation, storage, rotation, separation of duties and revocation for cryptographic keys and signing material.

Periodic review of who has access to production, timely removal on role change or departure, and administrative account inventory.

A written plan with roles, contacts, severity definitions, customer notification rules and evidence of at least one exercise or real incident review.

Intake of external reports, triage, severity assignment, remediation deadlines and communication back to the reporter.

Changes to production are reviewed, tested, approved and can be rolled back; emergency changes are recorded.

Inventory of sub-processors and critical vendors, their access, contractual security obligations and periodic review.

Defined retention periods, working deletion on customer request, deletion from backups within a declared window.

Direct and indirect prompt injection through user input, retrieved documents and tool outputs; separation of instructions from data.

Every action an agent can perform is bound to the permissions of the user it acts for, with confirmation for irreversible operations.

Model output is treated as untrusted: rendering, code execution, file and network access are constrained.

System prompts, other users' data and secrets cannot be extracted through the model; context is scoped per user.

Retrieval sources respect the requesting user's permissions; training or fine-tuning data is governed and free of customer secrets.

Provenance of models, weights, prompts and plug-ins; version pinning and change review.

Limits on token usage, tool invocations and cost per user; detection of automated abuse.

Agent actions are logged and attributable; high-impact actions have a human checkpoint; users can see what was done on their behalf.