Compliance & Enterprise Usage

This document describes how vigil aligns with compliance frameworks and how to integrate it in enterprise environments that use AI agents to generate code.

Why vigil in an enterprise environment

Organizations that adopt AI agents (Copilot, Cursor, Claude Code, ChatGPT) to generate code face new risks that traditional SAST tools do not cover:

Supply chain via hallucinations: Agents invent package names. An attacker can register that name with malware.
Example secrets in production: Agents copy values from documentation ("your-api-key-here") that end up in production.
Auth without controls: Agents generate functional endpoints but without authentication middleware.
Cosmetic tests: Agents generate tests that pass but do not verify anything real.

vigil detects these 4 patterns in a deterministic, auditable manner without dependency on external AI APIs.

Alignment with compliance frameworks

OWASP Top 10 for LLM Applications (2025)

vigil aligns directly with 3 categories of the OWASP Top 10 for LLM Applications:

OWASP Category	vigil rules	Coverage
LLM02 — Sensitive Information Disclosure	SEC-001, SEC-002, SEC-003, SEC-004, SEC-006, AUTH-004	Detects hardcoded secrets, placeholders, connection strings with credentials
LLM03 — Supply Chain Vulnerabilities	DEP-001, DEP-002, DEP-003, DEP-005, DEP-007	Detects hallucinated dependencies, typosquatting, suspiciously new packages
LLM06 — Excessive Agency	AUTH-001, AUTH-002, AUTH-005, AUTH-006	Detects endpoints without auth, permissive CORS, insecure cookies

EU Cyber Resilience Act (CRA)

The CRA requires that products with digital components be secure “by design”. vigil contributes to:

CRA Requirement	How vigil contributes
Vulnerability management in third-party components	DEP-001 through DEP-007 verify that dependencies exist, are legitimate, and are not typosquatting
Protection of stored and in-transit data	SEC-001 through SEC-006 detect hardcoded credentials that compromise data
Adequate access control	AUTH-001 through AUTH-007 detect endpoints without authentication and permissive configurations
Adequate testing	TEST-001 through TEST-006 detect tests that verify nothing (since v0.5.0)

SOC 2 Type II

Trust Service Criteria	Relevant vigil rules
CC6.1 — Logical access controls	AUTH-001, AUTH-002, AUTH-005, AUTH-006
CC6.6 — External threats	DEP-001, DEP-002, DEP-003 (supply chain)
CC6.7 — Credential management	AUTH-004, SEC-001, SEC-002, SEC-003, SEC-004, SEC-006
CC7.1 — Vulnerability management	All DEP- rules for supply chain

ISO 27001:2022

Control	Relevant vigil rules
A.8.25 — Secure development lifecycle	Integrate vigil into CI/CD as a quality gate
A.8.26 — Application security requirements	AUTH-001 through AUTH-007 verify access controls
A.8.28 — Secure coding	SEC-001 through SEC-006 detect secrets in code

NIST Cybersecurity Framework (CSF) 2.0

Function	Category	vigil rules
Identify	Asset Management	DEP-001 through DEP-007 (dependency inventory)
Protect	Access Control	AUTH-001 through AUTH-007
Protect	Data Security	SEC-001 through SEC-006
Detect	Continuous Monitoring	Integrate vigil into CI/CD pipelines

Integration in enterprise pipelines

Recommended pipeline

AI-generated code
    |
    v
[1. vigil scan]          <- Detects AI-specific patterns
    |
    v
[2. Semgrep/Bandit]      <- General-purpose SAST
    |
    v
[3. Snyk/Dependabot]     <- CVEs in dependencies
    |
    v
[4. Gitleaks]            <- Real leaked secrets
    |
    v
[5. Tests + Coverage]    <- Functional quality
    |
    v
Deploy

vigil complements (does not replace) existing tools. Its unique value lies in detecting problems specific to AI-generated code.

Strategy by environment

Environment	Strategy	`--fail-on`	Notes
Local development	`relaxed`	`critical`	Fast feedback without blocking
CI/CD (feature branch)	`standard`	`high`	Balance between speed and security
CI/CD (main/release)	`strict`	`medium`	Maximum rigor before production
Security audit	`strict`	`low`	Complete report for analysis

GitHub Actions example (production)

- name: Security scan (AI-specific)
  run: |
    pip install vigil-ai-cli
    vigil scan src/ \
      --format sarif \
      --output results/vigil.sarif \
      --fail-on medium \
      --config .vigil.strict.yaml

- name: Upload SARIF
  uses: github/codeql-action/upload-sarif@v3
  with:
    sarif_file: results/vigil.sarif

SARIF format for security platforms

vigil produces reports in SARIF 2.1.0, compatible with:

GitHub Code Scanning (native)
GitLab Security Dashboard
Azure DevOps
SonarQube (via importer)
Defect Dojo
Snyk (via SARIF importer)

Each finding includes ruleId, level, location, and optionally fixes with fix suggestions.

Reports for audits

Generate complete JSON report

vigil scan src/ --format json --output audit-report.json

The JSON report includes:

findings: Complete list of findings with rule_id, severity, message, location, suggestion
findings_count: Total number of findings
files_scanned: Files analyzed
duration_seconds: Execution time
analyzers_run: Analyzers executed
errors: Errors during analysis (if any)
version: vigil version used

Generate SARIF report with CWE references

vigil scan src/ --format sarif --output audit-report.sarif

The SARIF report maps each rule to CWEs when applicable:

CWE	Name	vigil rules
CWE-306	Missing Authentication for Critical Function	AUTH-001
CWE-208	Observable Timing Discrepancy	AUTH-007
CWE-614	Sensitive Cookie Without ‘Secure’	AUTH-006
CWE-798	Hard-coded Credentials	AUTH-004, SEC-001, SEC-002, SEC-003
CWE-829	Inclusion of Untrusted Functionality	DEP-001, DEP-003
CWE-862	Missing Authorization	AUTH-002
CWE-942	Permissive Cross-domain Policy	AUTH-005

Privacy and security of vigil

vigil is designed to be safe in enterprise environments:

No telemetry: Does not send data to external servers.
Deterministic: Does not use AI, ML, or external inference APIs.
Limited HTTP: Only makes GET requests to public PyPI/npm APIs to verify package existence. Disabled with --offline.
Local cache: Registry responses are cached at ~/.cache/vigil/registry/ (they do not contain project data).
Auditable: Each rule is a defined pattern in code. No black boxes.
No side effects: vigil only reads files. It never modifies code, sends messages, or creates resources.

Air-gapped mode

For environments without internet access:

vigil scan src/ --offline

In offline mode, only static checks are executed (typosquatting by similarity, auth patterns, secrets). Rules that require registry verification (DEP-001, DEP-002, DEP-005, DEP-007) are automatically skipped.

Enterprise FAQ

Does vigil replace Semgrep/Snyk/SonarQube?

No. vigil detects patterns specific to AI-generated code that other tools do not cover. It is complementary:

Problem	vigil	Semgrep	Snyk	Gitleaks
Hallucinated package (does not exist)	Yes	No	No	No
Typosquatting	Yes	No	Partial	No
Placeholder secret (`"changeme"`)	Yes	Partial	No	No
Real leaked secret	No	No	No	Yes
Known CVE in dependency	No	No	Yes	No
SQL injection	No	Yes	No	No

How long does a scan take?

vigil is fast because it does not perform AI inference. A typical scan:

100 files: < 1 second (offline), < 5 seconds (with registry verification)
1000 files: < 5 seconds (offline), < 30 seconds (with cold cache)

Can it be used in monorepos?

Yes. vigil supports multiple paths and file discovery performs automatic pruning of directories like node_modules/, .venv/, etc.

vigil scan services/auth/ services/api/ libs/shared/

How to exclude false positives?

Two options:

Disable the rule for the entire project:

rules:
  AUTH-005:
    enabled: false

Change the severity so it does not block:

rules:
  AUTH-005:
    severity: "low"