August 2, 2026: the EU AI Act deadline

The EU AI Act (Regulation (EU) 2024/1689) is the world’s first binding AI legislation. Published in August 2024, it’s being rolled out in phases. The goal is straightforward: regulate AI systems based on the risk they pose to people. The bigger the risk, the stricter the rules. The law applies to anyone using AI within the EU, even if you didn’t build that AI yourself. Got a subscription to a contact center platform with AI built in? You’re responsible for what gets communicated to customers in your name.

Spoiler: odds are, this applies to your organization too.

On August 2, 2026, the transparency obligations under Article 50 become fully enforceable. In short: if you’re putting AI in front of customers, you need to be upfront about it. Article 50 breaks that down into three obligations.

1. Your bot has to introduce itself
AI chatbots and voicebots must clearly state, right from the first interaction, that the customer is talking to an AI. Not buried in the terms and conditions, not tucked away at the bottom of a page. Think of an opening line like: “Hi, I’m [company]’s virtual assistant. How can I help you?” That’s the bare minimum.

2. AI-generated content has to be recognizable
AI-generated content covers any text, audio, or video that’s fully or largely produced by AI. Automated emails, chat replies, informational posts, that kind of thing. Customers need to know when they’re receiving that kind of content. The technical labeling requirement (a machine-readable label, like a watermark) has been pushed to December 2, 2026, but the obligation to actively inform customers already kicks in on August 2.

3. Voice analysis: two levels, two rules
This distinction matters:

• Level 1: Text-based sentiment analysis. AI that classifies a conversation or message as positive, neutral, or negative. This is allowed and falls outside the strictest rules.
• Level 2: Emotion recognition from voice. AI that analyzes intonation, speech rate, or tone of voice to infer emotions. This falls under Article 50 and is only permitted with an active notice to the customer, and in many cases, not permitted at all without a compelling reason.

This is the misconception that could cost companies dearly. The law doesn’t only apply to tech companies. Anyone using AI in their business has to comply, whether or not they built that AI themselves.

Under the law, that makes you a deployer. Using an external contact center platform with a built-in chatbot or AI-based routing? Article 50 applies to you. Transparency toward your customers is your responsibility, regardless of what your supplier has arranged.

The heavier obligations for high-risk AI have been pushed to December 2, 2027, through the Digital Omnibus agreement. But that delay explicitly does not cover Article 50. Anyone hoping for breathing room on their chatbot or voicebot: that room doesn’t exist.

Three concrete obligations came up above: your bot has to introduce itself, AI-generated content has to be recognizable, and you need to know what your voice analysis is actually doing. But knowing is one thing, having it sorted is what counts on August 2. Here’s the checklist:

☐ Take stock of your AI use in customer contact. Which channels run on AI? Chatbot, voicebot, email automation, conversation routing? Write it down.

☐ Check the opening message of every bot. Does it explicitly state, right at the first interaction, that the customer is talking to an AI? If not, this is the quickest and most important fix.

☐ Check whether your platform uses sentiment analysis or emotion recognition. Text classification is fine. Voice analysis falls under stricter rules. Ask your supplier explicitly which level is active in your environment, and get it documented.

☐ Ask your platform supplier for Article 50 documentation. Which AI features are active in your environment? Request it in writing and keep it on file.

☐ Document AI literacy within your team. Article 4 is a best-efforts obligation, document who received what training. It’s not a one-time event. Enforcement also starts on August 2.

Violations can lead to fines of up to €15 million or 3% of global turnover. But honestly, the reputational damage is worse. A customer who finds out, after the fact, that they were talking to an AI without ever being told, may not stay a customer.

Deploying voice AI while staying compliant isn’t a side note for our customers. From a voicebot’s opening message to how conversation data gets stored: compliance is baked into the way we implement.

Want to understand what the AI Act actually involves first? Read our explainer on its impact on customer service.