EU AI Act — Enforcement: 2 August 2026

EU AI Act compliance, verified by code.

Sentinel runs a deterministic 18-phase forensic scan of your repository and maps every detected signal directly to EU AI Act articles. Cryptographically signed. Always reproducible.

Start free audit See sample report

Static analysis only — zero code execution, zero side effects

RSA-PSS signed · SHA-256 verified · RFC8785 canonical

Annex IV technical documentation generated automatically

Compliance score

72/100

Aligned

Critical · Art. 14

No human override mechanism detected

Dossier ready

7 outputs generated

Annex IV · SARIF · SBOM · +4

app.gettingsentinel.com/audits/AUD-2026-04721

GAPAUD-2026-04721

TalentScreen Pro v3.1

HIGH_RISK · Annex III §4 · HR/Employment

Article Coverage

Art. 9

84%

Art. 14

31%

Art. 10

89%

Art. 20

96%

Art. 15

58%

Top Findings

CriticalArt. 14No human override mechanism detected

HighArt. 10Data governance documentation incomplete

MediumArt. 15Adversarial robustness tests absent

audit.jsonSARIF v2.1Annex IVSBOMRSA-PSS

0EU AI Act articles

0Pipeline phases

0+Signal patterns

0Output formats

0Languages

The process

From repo to signed audit in minutes

No manual configuration. No ambiguity. Identical input always produces identical output.

Step 01

Connect your repository

Add a sentinel.manifest.json. Sentinel reads your code with static analysis — nothing is executed, nothing changes in your system.

GitHub · GitLab · Bitbucket · any URL

Step 02

18-phase forensic scan

AST parsing across 15 languages, 89 AI package signatures, contradiction engine — every finding mapped to an exact file and article.

~5 min average · fully deterministic

Step 03

Receive signed audit bundle

Score, article breakdown, findings, Annex IV pack — RSA-PSS signed, SHA-256 verified, reproducible.

7 output formats included

Full technical breakdown

What gets assessed

22 articles scored.
Every signal traced to code.

Sentinel maps 33+ technical patterns directly to EU AI Act obligations. Each finding references the exact file, function, and line.

Art. 9Risk Management

84%11pt

Art. 14Human Oversight

47%11pt

Art. 10Data Governance

89%8pt

Art. 20Logging & Traceability

96%8pt

Art. 5Prohibited Practices

100%7pt

Art. 15Robustness & Security

58%5pt

22 articles scored · ARTICLE_WEIGHTED_V3

Risk categories

PROHIBITEDArt. 5

Hard FAIL — any prohibited practice overrides all other scores regardless of result.

HIGH RISKAnnex III

Full 22-article assessment. All Annex IV documents required. Sector multipliers applied.

LIMITED RISKArt. 50

AI output transparency obligations. Disclosure patterns verified in user-facing code.

MINIMAL RISKUniversal

Universally applicable articles only. Lighter assessment scope.

GPAIArt. 53

General-purpose AI obligations. Separate scoring track for model providers.

Scoring model

Four verdicts. Four tracks.

A numeric score alone doesn't determine compliance. Four independent evaluation tracks run in parallel — any can override the final verdict.

Verdict thresholds

≥ 85

Statically Aligned

All static signals satisfy requirements

≥ 65

Aligned

Material compliance achieved, minor gaps

≥ 40

Gap

Material gaps — remediation plan generated

< 40

Fail

Hard violations — fundamental remediation required

Override tracks

Governance Track

Required documents must exist and exceed quality threshold. Missing Annex IV overrides the score.

Technical Track

Code-level signals for critical articles verified independently. A failing article can't hide in an aggregate.

Score Gate

Hard floor per article — no single article may fall below its minimum regardless of others.

SIG Integrity

Comments stripped before verification. Compliance written only in comments is detected and rejected.

Output formats

audit.jsonreport.htmlSARIF v2.1.0Annex IVcompliance_report.mdepistemic_map.jsonremediation_roadmap

Use cases

Built for every compliance role

AI Providers

Audit before market placement. Generate all Annex IV documentation, gap report, and EU Declaration of Conformity evidence pack — ready for a Notified Body.

Pre-market compliance audit

Annex IV documentation pack

Gap remediation plan

Compliance Teams

Run audits on every release. Catch regressions before they reach the regulator. SARIF integrates with your CI/CD and GitHub Security tab.

CI/CD pipeline integration

Delta reports per release

Regression detection

Legal & Consultants

Deliver evidence-based assessments to clients. Signed, reproducible, legally formatted — credible in front of any regulator or Notified Body.

Client portfolio audits

Signed evidence bundles

Notified Body ready pack

CI/CD Integration

Plug into your pipeline. Zero configuration.

SARIF v2.1.0 integrates natively with GitHub Security tab, GitLab SAST, VS Code, and any CI/CD pipeline. Compliance on every commit.

GitHub Actions · GitLab CI · Bitbucket Pipelines

GitHub Security tab — native SARIF upload

VS Code Sentinel extension — findings inline

Delta report: what changed since last audit

Full integration docs

.github/workflows/sentinel.yml

# Sentinel EU AI Act audit — add to any workflow

- name: Run Sentinel audit

  uses: sentinel-ai/audit-action@v2

  with:

    token: ${{ secrets.SENTINEL_API_KEY }}

    upload-sarif: true

    fail-on: critical

# Outputs: audit.json · report.html · SARIF · Annex IV

# Results appear in GitHub Security → Code scanning

✓ Audit complete · Score 72/100 · 3 critical findings · SARIF uploaded

EU AI Act enforcement

2 August 2026.
Non-compliance is not an option.

High-risk AI systems must hold full technical documentation, human oversight mechanisms, and risk management evidence — or face fines of up to €30M or 6% of global annual turnover.

Feb 2025

GPAI obligations in force

Aug 2025

Prohibited practices ban active

Aug 2026

HIGH-RISK systems full obligationsNOW

Aug 2027

Notified Body conformity assessments

Start audit before deadline

FAQ

Common questions

Enforcement: 2 August 2026

Start your EU AI Act audit today.

Connect your repository and get your compliance score in minutes. Signed, reproducible, legally formatted.