You use ChatGPT at work. Here's what the EU AI Act actually asks of you.

You're in Slack. Someone drops a link to an EU AI Act article. Your stomach drops. You use ChatGPT daily. You have no idea if you're covered. Fair. We'll sort it.

Here's the short answer for you

You show EU customers AI output — a chatbot, auto-drafted emails, anything they see. Article 50 kicks in from August 2, 2026. You need a disclosure.

You only use ChatGPT internally — summaries, drafts, code — Article 4 is your main job. Your team needs basic AI literacy. You need proof.

Neither is hard. You still need to handle both.

What the EU AI Act is (in one minute)

The EU AI Act (Regulation 2024/1689) is Europe's AI law. It sorts AI by risk. Higher risk means more rules.

It started August 2024. Phased rollout began February 2025.

You're UK-based? It still hits you if you have EU customers, EU staff, or EU sales. Same logic as GDPR. Brexit doesn't shield you from EU market rules.

What risk bucket does your ChatGPT use fall into?

ChatGPT is a General Purpose AI model. Your risk depends on how you use it — not the tool itself.

You use ChatGPT only internally (drafting, summarising, research): Minimal risk. Article 4 AI literacy applies. Example: your team summarises meeting notes in ChatGPT. Nobody outside the firm sees the output.

You use ChatGPT where customers see it (chatbots, auto-replies, public posts): Limited risk. Article 50 transparency applies. Example: your Intercom bot drafts replies from ChatGPT. EU visitors need to know it's AI.

You use ChatGPT for decisions that seriously affect people (hiring, credit, healthcare, policing): Potentially high risk. Much heavier rules. Most small teams aren't here. Example: you auto-screen CVs with ChatGPT and reject candidates without human review.

What Article 50 actually requires from you

Article 50 is the transparency rule. For most ChatGPT users, it's the big one before August 2, 2026.

Plain English version:

1. Customers interact with your AI? Tell them it's AI. Put it on your Zendesk bot. Put it in your website chat widget. Before the conversation starts.

1. You publish AI-generated content on public-interest topics? Label it as AI-generated. Example: AI-written policy explainers on your blog.

1. Your system reads biometrics or detects emotions? Tell affected people. Example: a retail kiosk that guesses shopper mood from facial data.

Most of you only need item one. A customer-facing ChatGPT bot needs one line: "You're chatting with AI." That's enough. Don't overthink it.

What Article 4 requires from your team

Article 4 is AI literacy. It applies to everyone using AI — customer-facing or not.

You deploy AI at work? Your staff need "a sufficient level of AI literacy." They need to know what AI can do. They need to know what it can't. They need to know the rules that apply.

In practice: run a short internal training session. Log who completed it. NudgeCompliant has five modules and generates completion records for you.

Your practical checklist

You're a UK team using ChatGPT. Here's what to do.

*By August 2, 2026:*

• Add a disclosure to every customer-facing AI touchpoint (chatbot, auto-email, public content)
• Get your team through AI literacy training — keep the completion log
• Write down which AI tools you use and what for (ChatGPT for proposals, Copilot for code, etc.)

*By December 2, 2027 (only if high-risk):*

• Most ChatGPT use stays out of high-risk categories
• Not sure? Run a formal risk assessment

*What you don't need to do:*

• Register your AI system — that's for providers, not you
• Stop using ChatGPT
• Hire a consultant — unless your use case is genuinely complex

Why this matters beyond fines

August 2, 2026 is [X days] away. A chatbot disclosure is one line of text. AI literacy is one training session with a log.

Regulators probably won't fine a 20-person agency on day one. But procurement teams will ask. Enterprise clients want AI governance docs before they sign. Investors check during due diligence. Non-compliance becomes a sales blocker fast.

Your 10-minute action: Open your customer-facing chat tool right now. Add one sentence: "This service uses AI." Save it. You're done with the hardest part of Article 50.

This article is for information only. It's not legal advice. For complex situations, talk to a qualified lawyer.