Skip to content
EU AI Act self-assessment for AI startups and SaaS

Get your EU AI Act position in writing.

When an enterprise vendor review or investor due diligence asks where your AI stands, answer with a dated report. It takes about 10 minutes, after a free 2-minute scan.

Get your risk tier free See a sample report

No card or account for the scan · One-time payment · Updated for the June 2026 Omnibus

Sample report for Kadra, a fictional product

When founders need this

Check your risk tier for free

Eight questions and an email step, about 2 minutes. No card, no account. Answers are never sold or used to train AI.

Art. 5 · Annex III · Art. 50 · Art. 4

Four risk tiers

The Act sorts AI systems by the risk of how they are used, and the tier sets your obligations and deadlines. Pick a tier to see what it means.

Unacceptable riskArt. 5

Practices the Act bans outright, because they are a clear threat to people's safety or rights.

Examples

  • Social scoring that leads to unfair treatment
  • Emotion recognition at work or in school
  • Scraping faces from the web to build recognition databases

What you must do

  • Do not offer or use it in the EU: no paperwork makes it lawful
  • If a feature comes close, have it reviewed and redesign it before launch

When Banned since 2 Feb 2025

Check yours in 2 minutes Read the full explanation

High riskAnnex III

AI used where a wrong output can seriously affect someone's life, work or safety. Allowed, with the Act's heaviest obligations.

Examples

  • Screening or ranking job applicants
  • Scoring a person's creditworthiness
  • Grading exams or deciding admissions

What you must do

  • Check the Article 6(3) exemption first: it may take you out
  • Providers: risk management, documentation, logging and human oversight
  • A conformity assessment and registration before launch

When From 2 Dec 2027 (Annex III)

Check yours in 2 minutes Read the full explanation

Limited riskArt. 50

AI that talks to people or generates content. The duty is mostly transparency: people should know AI is involved.

Examples

  • A customer support chatbot
  • An AI voice agent on the phone
  • AI-generated images, audio or video

What you must do

  • Tell people they are talking to an AI, unless it is obvious
  • Mark AI-generated content in a machine-readable way (providers of generative AI)
  • Disclose deepfakes you publish

When Disclosure since 2 Aug 2026, marking from 2 Dec 2026

Check yours in 2 minutes Read the full explanation

Minimal riskArt. 4

Most AI in use today. No obligations specific to this tier, beyond AI literacy.

Examples

  • Spam filters
  • Product recommendations
  • Most internal tools, such as document search

What you must do

  • Make sure staff who build or use it have enough AI literacy (Art. 4)
  • Check again if the use changes, for example if it starts deciding about people

When AI literacy since 2 Feb 2025

Check yours in 2 minutes Read the full explanation

One system can sit in more than one tier: a hiring chatbot can be high-risk and still owe the Article 50 disclosure. General-purpose AI models follow a separate track (Art. 53).

The June 2026 Omnibus

What applies now, and what can wait

The Omnibus deferred the high-risk obligations. It did not move the transparency duties, and regulators can already fine for those.

In force now
Feb 2025

Prohibited AI practices (Article 5) and staff AI literacy (Article 4). Enforced since 2 Aug 2026.

Not delayed
2 Aug 2026

Article 50(1): chatbots must tell people they are talking to an AI. Regulators can fine from this date; for SMEs the cap is the lower of €15M or 3% of turnover (Art. 99(6)). Labelling AI-generated content follows on 2 Dec 2026.

Deferred, not cancelled
2 Dec 2027

High-risk obligations (hiring, credit, education, health, infrastructure AI): risk management, documentation, conformity assessment, CE marking.

Free tools

Quick checks for the questions that come up first.

Art. 50Article 50 disclosure generatorWording that tells users they are dealing with an AI system. Annex IIIHigh-risk checkerTest a use case against Annex III and the Article 6(3) exemption. Art. 99Fine calculatorThe penalty caps for your company size, SMEs included.
How it works

Three steps. No lawyers yet.

No account, no sales call, no 40-page PDF of legal boilerplate.

01

Answer the assessment

What your AI does, who it affects and how it is governed. Up to 33 questions, or 41 with the Document Pack, in about 10 minutes.

02

AI builds your gap report

Your answers are analysed against the current, post-Omnibus EU AI Act. Every section references your actual product - not generic boilerplate.

03

Get your fix list by email

Risk tier, the Articles that apply to your answers, a deadline per obligation, and a prioritised remediation checklist you can hand to your team - or your lawyer.

What's inside

Everything a first call with your lawyer would start with.

Written for founders and product managers, not lawyers.

Your risk classification

Unacceptable / High / Limited / Minimal - in plain English, with why.

Obligations + post-Omnibus deadlines

The Articles that apply to your answers, each with its date as amended by the Omnibus - not last year's.

Prioritised fix list

What to fix first, and why.

GDPR overlaps

Where one answer satisfies two regulations.

What you're doing right

It's not all gaps.

The exact questions for your lawyer

So that conversation starts at question 10, not question 1. Stamped with the regulation status after the June 2026 amendments.

Optional: Document Pack

3 draft compliance documents pre-filled from your answers.

See a sample report

Pricing

Pay once. That's it.

A dated report you can share with a buyer's procurement team, an investor or your lawyer.

One AI system

For a single product or AI feature.

See a sample report first →

Full Gap Report

Report only

€59
one-time · no subscription

Your position in writing, for your team and your lawyer.

  • Full branching assessment (~10 min)
  • Provider vs deployer role analysis
  • Article 6(3) exemption check
  • GPAI / Article 53 check
  • Obligations + post-Omnibus deadlines per tier
  • Prioritised remediation checklist
  • Trust badge · Notion tracker · 12-month re-issue
Get your report
Recommended

Deal-Ready Pack

Report + Document Pack

€149
one-time · no subscription

The report plus three drafts you can adapt for a vendor review.

  • Everything in the Full Gap Report
  • Article 50 transparency notice (draft)
  • Human oversight procedure (draft)
  • Risk management plan (draft skeleton)
  • All pre-filled from your answers
  • Up to 41 questions, so the drafts fit your system
Get the Deal-Ready Pack

More than one AI system?

Each system gets its own full report, because each has its own risk tier.

3 Gap Reports

Report only, for 3 systems

€149
≈ €50 per report
  • A full gap report for each system
  • 12-month re-issue for every system
  • Portfolio summary of all 3 systems, within 3 working days of your third report
  • Use the reports in your own product or client work
  • Without the Document Pack
Get 3 reports

3 systems, Deal-Ready

For agencies, fractional DPOs and portfolio operators

€399
€133 per system
  • A full gap report and the three drafts for each system
  • 12-month re-issue for every system
  • Portfolio summary of all 3 systems, within 3 working days of your third report
  • Use the reports in your own product or client work
  • Your firm's logo on the report covers (upload it in the assessment)
Get the 3-system pack

Enterprise

More than 3 systems

Custom quote
Volume pricing, quoted by email
  • Gap Report or Deal-Ready Pack for each system
  • Tell us how many systems and which version you need
  • The same automated reports and drafts as the standard products
  • White-label reports available on request
Contact us

Free risk scan Eight questions · no card Your likely tier, before you spend anything.

Start free scan

Prices exclude VAT, added at checkout where applicable. 14-day refund.

The regulation moves. Your report keeps up. The EU changed the AI Act's deadlines once already this year. If it materially changes again within 12 months of your purchase, we re-run your assessment and send you an updated report - free, once, no questions.

FAQ

Questions founders actually ask

Didn't the EU just delay the AI Act?

Partly. The June 2026 Omnibus moved high-risk obligations to 2 Dec 2027, and to 2 Aug 2028 for AI in regulated products. Chatbot disclosure under Article 50(1) has applied since 2 Aug 2026, the day regulators gained fining powers, and labelling AI-generated content follows on 2 Dec 2026. The report uses the current dates, obligation by obligation.

Does the EU AI Act apply to companies outside the EU?

Often, yes - incorporation is not the test. Article 2 reaches providers irrespective of where they are established, and reaches providers and deployers in third countries where the output produced by the AI system is used in the Union. A US or UK company with EU users is normally in scope. But being in scope is not the same as being high-risk: most non-EU SaaS lands at limited or minimal risk. Full territorial scope guide.

Which AI model does Reglynn use?

Reports are generated by an AI system built on Anthropic's Claude, running Reglynn's own question set and risk logic, and delivered automatically. Reglynn is itself a Limited Risk AI system under Article 50 and discloses AI involvement wherever it occurs, including in our chat assistant. Our own self-assessment.

Why not just ask ChatGPT?

You can - but it only answers the questions you know to ask. The question set is the product: it surfaces contestability mechanisms, DPIA requirements, logging duties, and literacy obligations most founders have never heard of. And a chat transcript isn't something you can show an investor or a due-diligence team. A structured report with your company name, tier, and fix list is.

Is this legal advice?

No. This is a structured self-assessment based on the EU AI Act (Regulation 2024/1689, as amended). It tells you where you likely stand and exactly what to verify with a qualified advisor - so that conversation is shorter and cheaper.

What happens to my answers?

Your answers and your report are kept so we can answer questions about them and re-issue the report if the regulation materially changes. They're never sold, shared beyond the processors listed in our privacy policy, or used to train AI models. Email hello@reglynn.eu and we will delete them.

What if my product is classified "Unacceptable"?

Then you need to know that today, not in an enforcement letter. The report covers exactly which provision applies, why, and what a compliant redesign path could look like.

I have more than one AI system.

Each system needs its own assessment, because each has its own risk tier: a chatbot and a scoring model are not the same regulatory object. The multi-system bundles under Pricing cost less per system and include a one-page portfolio summary of all 3 systems. Each report is emailed automatically within minutes of finishing that system's assessment. Ask for the summary in the last step of the assessment (on the 3-system Deal-Ready bundle you also upload your logo for the report covers there), and we send it within 3 working days of your third report. You may use the reports in your own product or client work. Run the assessment once per system; each gets its own full report. For more than three systems, email us for a quote; white-label reports are available on request.

Procurement will ask where your AI stands.Have the answer in writing.

Eight questions, free, no card. Your risk tier is on screen in about 2 minutes.

Get your risk tier free See a sample report

Legal disclaimer: This tool provides a self-assessment only and does not constitute legal advice. Classifications are based on publicly available guidance, including the EU AI Act (Regulation 2024/1689) as amended by the June 2026 Omnibus. For binding determinations, consult a qualified EU AI Act compliance advisor.

Data handling: Submitted answers and the resulting report are kept as a delivery and quality record. No answers are sold, shared beyond our listed processors, or used for AI training, and we delete them on request.