EU AI Act · Regulation (EU) 2024/1689In force since 2 Aug 2026

The EU AI Act doesn’t scale down.Your compliance budget shouldn’t have to.

Article 50 transparency and GPAI model obligations are enforceable now. Fines reach €15 million or 3% of worldwide annual turnover, whichever is higher. The law applies to anyone who puts their name on an AI system that talks to people or generates content — not just the big model labs.

€15M is a rounding error for one company, an extinction event for another. That asymmetry is what ThinkNEO corrects.

Start governing — from $19/moSee how it worksDirect self-serve signup. No form, no sales call.
ThinkNEO · control planelive
→ incoming
POST /v1/chat/completions
# intent: Support chatbot answer
# policy: Art. 50 · AI-interaction disclosure
Verdict
LABEL
prepend AI-disclosure notice
Every AI request through ThinkNEO is evaluated against your policy set before the model answers.
Quick check · 30 seconds

Am I affected by the EU AI Act?

Two questions. If both are yes, Article 50 obligations apply to you today.

01

Do you sell to or serve customers in the EU?

Full assessment · 3 minutes

Check your EU AI Act readiness in 3 minutes

Two separate outputs. First: does the law apply to you, and how. Second: what operational controls you’re missing.

No email required for the result.

The 4 obligations most SMBs miss

Article 50 in plain English

In force since 2 Aug 2026. Applies to anyone putting their name on an AI system in the EU market, not just model labs.
01

AI that interacts with people

Article 50 · in force · subject to fines

Chatbots, voice assistants, in-app AI agents — including a branded support bot you built on top of someone else’s model.

ThinkNEO

ThinkNEO applies the disclosure rule at runtime, on every in-scope request. Not a note in your docs — a policy the control plane enforces before the model answers.

02

Synthetic content — marking & detection

Article 50 · in force · subject to fines

Any image, audio, or video generated or substantially manipulated by AI, unless a substantive human editorial layer intervenes.

ThinkNEO

ThinkNEO embeds C2PA-compatible markers and keeps a registry entry per artifact. Detection stays functional even after re-encoding.

03

Deepfake disclosure

Article 50 · in force · subject to fines

Content depicting real people in situations that didn’t happen — unless part of clearly artistic, satirical, or public-interest work with disclosure.

ThinkNEO

ThinkNEO detects deepfake generation attempts, requires an editorial-review flag, and refuses if it’s absent.

04

Emotion recognition & biometric categorization

Article 50 disclose · Article 5 prohibits some uses

Systems that infer emotion, feelings, or characteristics from faces, voices, or biometrics must inform affected people (Article 50). In workplace or education contexts, these uses are PROHIBITED entirely under Article 5, since 2 Feb 2025.

ThinkNEO

ThinkNEO gates biometric-inference calls with a disclosure requirement and refuses uses that fall under Article 5’s prohibited list.

Prepares you for 2027+

High-risk systems phase in later.

If you operate in an Annex III area (employment decisions, credit scoring, education, critical infrastructure, migration, essential services), full high-risk obligations (Article 6, Annex IV technical documentation, human oversight evidence, risk monitoring, post-market surveillance) become enforceable in 2027-2028. ThinkNEO stores the evidence you’ll need — we don’t fake the certification today.

What the law asks · what ThinkNEO ships

Seven required controls, one control plane

ThinkNEO is a tool for deployers — the companies using AI systems, not the model labs that build them. Where an article primarily binds the provider, we cite the corresponding deployer duty (mostly Art. 26) so you can see which obligations you inherit and which you’d need from the model provider. The requirements are real and the work is real; below is what building each control yourself looks like versus what ThinkNEO ships out of the box.

EU AI Act needWithout ThinkNEOWith ThinkNEO
Immutable audit trail
Article Art. 12 (provider) · Art. 26(6) (deployer)
Custom append-only logs per provider. Manual verification.
Hash-chained audit ledger. One line per event. Verify anytime.
Per-request logging
Article Art. 12 (provider) · Art. 26(6) (deployer)
Provider SDK logs, disparate formats, no cross-provider view.
Unified request log, all providers, one schema.
Access control
Article Art. 26
Manual IAM, per-tenant custom code, drift on rotation.
Per-key policies, per-user scopes, rotate without downtime.
Policy enforcement
Article Art. 50
Middleware you maintain. Provider-specific. Bypassed by any new endpoint.
Runtime policies enforced before the model answers. New providers inherit the policy set.
Cost accountability
Article — (business practice, supports Art. 26)
Provider bill reconciliation. Manual tagging. Weekly finance reviews.
Per-system, per-team spend in real time. Alerts before overrun.
Runtime governance
Article Art. 26 · Art. 50 (deployer duties)
Documented process. Reviewed at audit time.
Enforced at request time. Evidence generated automatically.
AI system inventory
Article Art. 26 (deployer duties)
Spreadsheets. Out of date within weeks.
Auto-registry: one entry per system, owner + purpose. Updates on first use.
Prepares you for 2027+

High-risk regime (phases in 2027-2028)

We don’t ship these as “live” today. We store the evidence you’ll need.

  • Human oversight evidence
    2027-2028

    Article 14 (provider design) · Article 26(2) (deployer implementation) — evidence that a natural person reviewed or could intervene. We log the reviewer and the decision.

  • Risk monitoring
    2027-2028

    Article 9 (provider) · Article 26(5) (deployer) — ongoing risk management and intended-purpose monitoring. We log every enforcement event as candidate evidence.

  • Post-market monitoring plan
    2027-2028

    Article 72 (provider post-market monitoring) · Article 26(5) (deployer intended-purpose monitoring) — plan required. Our audit ledger is the raw feed you’ll build the plan from.

Pricing

Compliance shouldn’t cost more than the AI it governs

Every paid tier ships the same product — full runtime enforcement, inline policy blocks, budget hard-stops, complete audit ledger. Higher tiers don’t unlock features. They give you more room. Monitor-mode observation is free forever.

Plan · Start
$19/mo
1,000 governed requests / month
For: Solo developer shipping a first AI feature.
  • Full runtime enforcement (inline policy blocks, budget hard-stops)
  • Complete audit ledger
  • 1 tenant · 1 user · 5 projects · 3 API keys
  • Monitor-mode observation free forever
Start free trial
Plan · Dev I
$19.90/mo
10,000 governed requests / month
For: Growing team wiring AI into more surfaces.
  • Everything in Start — same product, more room
  • 1 tenant · 5 users · 15 projects · 10 API keys
Start Dev I
Plan · SMB I
$75/mo
100,000 governed requests / month
For: SMB running multiple products under real compliance scrutiny.
  • Everything in Dev I — same product, more room
  • 3 tenants · 50 users · 50 projects
Start SMB I
Plan · SMB II
$145/mo
500,000 governed requests / month
For: Mid-market operating several product lines.
  • Everything in SMB I — same product, more room
  • 10 tenants · 100 users · 100 projects
Start SMB II
Plan · Enterprise
Custom
Contract-set volume
For: Regulated organisations needing bespoke SLA, procurement, DPA.
  • Everything in SMB II — on contract terms
  • Bespoke SLA and procurement
Talk to sales

Prices in USD, billed monthly via Stripe. Included-request allowances degrade to observe-only (never a hard cutoff); structural limits (tenants, users, projects, keys) block creation at the cap. No metered overage.

FAQ

Common questions

Does the EU AI Act apply to me if I’m outside the EU?

Yes if the output of your AI system is used in the EU, or if you offer it to EU users. Physical location of your company doesn’t matter — market location does.

I just use OpenAI/Anthropic. Am I still subject to Article 50?

If you put your name on the system (branded chatbot, in-app assistant, generated content published under your brand), yes. Using someone else’s model doesn’t make you a provider — but re-exposing it to end users under your name makes you a deployer with Article 50 duties.

What are the fines really?

€15 million or 3% of worldwide annual turnover, whichever is higher, for most Article 50 breaches. €35M or 7% for prohibited uses under Article 5. National authorities enforce; enforcement modalities are being finalized.

Do I need a lawyer to start using ThinkNEO?

No. You need enforcement — the operational controls that prove you did what the law asks. That’s what we ship. Legal review comes when your Legal team drafts your public disclosure text or reviews your Annex-IV documentation.

How is ThinkNEO different from a compliance consultancy?

A consultancy tells you what the law says. We give you the runtime plumbing that satisfies it — logs, marks, disclosures, enforcement points, evidence — from $19/mo, without a form.

Governance shipped in an afternoon. Enforcement live before your next EU customer signs up.

You do not need a lawyer or a Big-Four consultant. You need the plumbing. That’s what we ship.