AI Compliance Firewall
EU AI Act enforcement. Before the AI responds.
SentinelAI stops non-compliant AI traffic before it reaches your AI provider — automatically. Change one URL in your code, or install the Desktop Client for zero-config ChatGPT & Copilot governance. 47 EU AI Act rules enforced in real time. 30+ sensitive data types anonymized. Every request logged. FRIA assessments, incident reports with legal deadlines, and compliance documentation — all built in.
The Problem
EU AI Act Compliance — Your AI Calls May Already Violate the Law
The EU AI Act is in force. Fines up to €35M are real. A regulatory audit can come at any time.
Fines up to €35 million
The EU AI Act imposes fines up to €35M or 7% of global revenue for prohibited AI practices. A single wrong API call can be costly.
Your ChatGPT call for HR violates Annex III
Using GPT for candidate screening? That's a high-risk AI system under Annex III §4. Human oversight is required — without it, you're in breach.
The regulator asks for an audit log — where’s yours?
Every AI call needs a trail: who sent it, what for, what data was involved, what the outcome was. Without that, compliance is impossible.
They also ask for your FRIA, incident reports, and EU registration proof
Beyond the audit log, deployers must produce a Fundamental Rights Impact Assessment (Art. 27), file incident reports with exact 15-working-day deadlines (Art. 73), and register in the EU AI database (Art. 49). Without a system, that’s a manual nightmare.
How It Works
Two Ways to Deploy — Same Compliance Engine
Choose the deployment that fits your workflow. Both options enforce 47 EU AI Act rules, anonymize 30+ PII types, and create a full audit trail.
Real-time screening
Every AI request passes through the rule engine: PII anonymization with Presidio NLP, risk classification, EU AI Act article matching.
→ UseCase: hr_screening
→ Risk: High (Annex III §4(a))
→ Outcome: pending_review
47 violation rules enforced
Complete EU AI Act coverage: all 8 Art. 5 prohibited practices blocked, all 8 Annex III high-risk areas monitored, Art. 50 transparency, GPAI obligations.
Annex III → 19 high-risk use cases
Art. 50 → 4 transparency rules
Art. 51/53/55 → GPAI obligations
GDPR Art. 6 → lawful basis check
Full compliance dashboard
Review flagged requests, manage FRIA assessments (Art. 27), file incident reports with auto-calculated deadlines (Art. 73), and track your 7-obligation compliance score (Art. 26).
FRIA #12 → Art. 27 → PDF export
Incident #3 → deadline: 15 working days
Compliance: 7/7 Art. 26 checks ✓
Proof
47 Violation Rules — Live Enforcement Examples
Real examples from the compliance engine. Every AI request is matched against EU AI Act articles, Annex III high-risk categories, and GDPR rules in real time.
Comparison
Real-Time Enforcement vs Post-Hoc Governance
Most AI governance tools audit after the fact. SentinelAI enforces before the AI responds.
| Category | SentinelAI | Traditional AI Governance |
|---|---|---|
| Enforcement | Real-time — before the AI responds | Post-hoc audits & dashboards |
| Deployment | API proxy or Desktop Client (zero-config) | SDK integration + months of setup |
| Prohibited AI blocking | Automatic HTTP 451 — request never reaches AI | Flagged in reports after the fact |
| PII protection | NLP + regex anonymization (30+ types) | Manual redaction or basic regex |
| Desktop AI apps | ChatGPT & Copilot governed via DNS proxy | Not supported |
| Legal coverage | 47 rules: EU AI Act + GDPR + UK + US | Questionnaire-based assessments |
| Deployer compliance tracking | 7 Art. 26 obligations scored; FRIA (Art. 27) workflow + PDF; incident reporting (Art. 73); EU database registration (Art. 49(2)) | Manual checklists |
| Audit trail | Automatic for every AI request | Requires manual logging setup |
| Time to production | ~15 minutes | 3–6 months |
Pricing
Simple, transparent
Start free. Scale as you grow.
- 1,000 requests / month
- EU AI Act coverage (47 rules)
- PII anonymization (30+ types)
- Basic audit log
- 1 user
- 10,000 requests / month
- EU + UK jurisdictions
- Human review workflow
- Compliance dashboard
- Up to 10 users
- Email support
- 150,000 requests / month
- EU + US + UK jurisdictions
- Desktop Client (ChatGPT & Copilot)
- Advanced compliance scoring (7 Art. 26 obligations)
- FRIA workflow — Art. 27 PDF export
- Incident reporting (Art. 73) — holiday-aware deadlines
- EU AI database registration (Art. 49(2))
- Up to 50 users
- Priority support
- Unlimited requests
- All jurisdictions
- On-premise deployment option
- Desktop Client with org-wide rollout
- Custom rule engine policies
- Unlimited users
- Dedicated account manager
Early Access
Reserve your free access
Be among the first to use SentinelAI. No obligations.
FAQ
Frequently Asked Questions About AI Act Compliance
Everything you need to know about EU AI Act compliance and how SentinelAI helps.
What is the EU AI Act and who does it apply to?
The EU AI Act (Regulation 2024/1689) is the world’s first comprehensive AI regulation, in force since August 2024. It applies to any company deploying or developing AI systems that affect people in the EU — regardless of where the company is based. This includes using AI APIs like OpenAI, Anthropic, or Azure OpenAI.
How does SentinelAI enforce AI Act compliance?
SentinelAI works as a real-time compliance firewall. It inspects every AI request against 47 EU AI Act rules before the request reaches the AI provider. Prohibited uses (Art. 5) are blocked with HTTP 451, high-risk uses (Annex III) are routed for human review, PII is anonymized with NLP, and a full audit trail is created automatically.
What is the Desktop Client and how does it work?
The Desktop Client is a lightweight Windows app that governs ChatGPT, Copilot, and other AI desktop applications without any code changes. It works via DNS-level interception — AI traffic is transparently routed through the same compliance engine that the API proxy uses. Your team installs the app, and all AI usage is governed. No browser extensions, no VPNs, no network redesign.
What AI practices does SentinelAI block?
SentinelAI enforces all 8 prohibited practices from Art. 5: social scoring, real-time biometric mass surveillance, subliminal manipulation, exploitation of vulnerable groups, predictive policing based solely on profiling, emotion recognition in workplaces/schools, untargeted facial recognition scraping, and law enforcement biometric identification (without authorization). These are blocked automatically — the request never reaches the AI provider.
What are the fines for EU AI Act non-compliance?
Fines for prohibited AI practices (Art. 5 violations) can reach up to €35 million or 7% of global annual revenue. High-risk system violations can result in fines up to €15 million or 3% of revenue. Even transparency violations (Art. 50) can cost up to €7.5 million or 1% of revenue.
Do I need compliance if I just use ChatGPT or GPT-4 APIs?
Yes. If you use AI APIs for high-risk purposes like HR screening, credit scoring, or legal decisions, you’re operating a high-risk AI system under Annex III of the EU AI Act. Human oversight (Art. 14), documentation (Art. 11), and risk management (Art. 9) are required — even if you didn’t train the model yourself.
How many PII types does SentinelAI detect?
Over 30 PII types, using Microsoft Presidio NLP/ML engine combined with custom regex recognizers. This includes names, emails, phone numbers, credit cards, IBANs, national IDs (JMBG, SSN), IP addresses, medical terms, passport numbers, and more — with support for Serbian and other regional formats.
What is a FRIA and does SentinelAI include one?
A Fundamental Rights Impact Assessment (FRIA) is required by EU AI Act Art. 27 for public sector bodies and many private operators deploying high-risk AI systems. SentinelAI includes a built-in FRIA module: create an assessment covering all 6 Art. 27(1) fields, submit for review, record the outcome (NoSignificantRisk / SignificantRisk / Mitigated), and download a PDF with a legal notice footer. Completed FRIAs are immutable — ensuring the assessment accurately captures the state of the system at the time of the original evaluation.
How long does it take to set up SentinelAI?
API integration: about 15 minutes. Change one URL in your code and add an API key header. Desktop Client: install the app and you’re done — ChatGPT and Copilot are governed immediately. No new SDKs, no refactoring, no infrastructure changes needed.