Skip to content
All posts

10 July 2026

5 min read

Written by

Clément Lacaille

Clément Lacaille

Founder, Tech-Bharat

About the author
Business & compliance

EU AI Act: chatbot compliance deadline August 2, 2026 — what SMBs must do

The EU AI Act’s chatbot transparency rules (Article 50) still take effect on August 2, 2026 — the delay to December 2027 only covers high-risk systems. Customer service chatbots, WhatsApp agents and AI assistants must disclose they are AI. Here is the SMB compliance checklist.

On 29 June 2026, the Council of the European Union gave its final approval to the “Digital Omnibus”, following the European Parliament’s vote on 16 June. The text postpones the AI Act’s toughest obligations: high-risk systems under Annex III — automated recruitment, credit scoring, school admissions, among others — no longer need to comply by 2 August 2026 but by 2 December 2027; Annex I systems, embedded in already-regulated products, gain until 2 August 2028. Signature was expected in early July, with publication in the EU’s Official Journal announced for the coming days, in time to enter into force before 2 August. Many SMB owners have concluded they now have until late 2027 to think about AI compliance. That’s wrong for anyone whose customers talk to an AI: Article 50 of the regulation, on transparency, is unaffected by this postponement and still applies on 2 August 2026.

What’s still due on 2 August 2026

  • Inform users they are talking to an AI, from the start of the conversation: a mention buried in the terms of service or a footer is not enough — you need an explicit message such as “You’re chatting with an AI assistant.”
  • Flag any deepfake content — AI-generated or manipulated image, audio or video that resembles a real person, place or event closely enough to look authentic — even for creative or satirical use, where a discreet disclosure is still mandatory.
  • Notify users if an emotion-recognition or biometric-categorisation system is in use.
  • Only the automatic technical marking (machine-readable watermarking) of content already in circulation gets extra time, until 2 December 2026.

What actually got postponed

The Digital Omnibus delay applies to high-risk systems strictly defined under Annex III: recruitment, credit scoring, school admissions, workforce management, access to essential public services. Most AI agents an SMB deploys today — invoice follow-up, customer support, sales qualification, reporting — do not fall into that category and were never on the hook for the 2 August deadline anyway. The risk is not missing a high-risk obligation you never had. It is assuming you are safe on transparency until 2027, when that particular obligation never moved.

What it means for your SMB

If your customer service or sales run through a chatbot, a WhatsApp sales agent, or any AI assistant that talks directly to your customers, the question to settle before 2 August is simple: does the very first message in the conversation clearly say it is an AI? In France, the CNIL is preparing to become the market surveillance authority for this regulation — non-compliance with the text’s obligations, outside prohibited practices, can be fined up to €15 million or 3% of global turnover. For an SMB, the real risk is not the maximum fine, rarely applied at that scale — it is an inspection turning up a chatbot script that was never audited.

Before August 2: four concrete checks

  • Map every customer-facing AI touchpoint — website chatbot, WhatsApp agent, voice assistant — and check that the very first message explicitly discloses it is an AI.
  • If you publish AI-generated or AI-edited visuals or videos that could pass as authentic, add a visible disclosure; a technical watermark alone is not enough for content already online before December 2026.
  • Do not rely on the Digital Omnibus delay if you do not use scoring, automated recruitment or biometrics — it changes nothing for you, unlike Article 50.
  • Document these checks in writing, with dated screenshots of the scripts in force — in an inspection, that trail is what separates a minor fix from a formal notice.

This is exactly the kind of deadline — a precise date, a narrow obligation, a text still moving right up to publication — that a regulatory watch agent is built to track on your behalf: reading official publications, isolating what actually applies to you, and flagging it before the deadline instead of leaving you to trust a headline that got the nuance wrong. 2 August 2026 is not the reassuring version of a postponed text. On transparency, it is a deadline that lands in three weeks.

Frequently asked questions

Does my customer service chatbot have to comply with the EU AI Act by August 2, 2026?+

Yes. Article 50 transparency obligations apply on 2 August 2026 and were not postponed by the Digital Omnibus. Any chatbot, WhatsApp agent or AI assistant that talks to your customers must disclose, from the very first message, that the user is interacting with an AI.

Wasn’t the AI Act delayed to December 2027?+

Only partly. The delay approved on 29 June 2026 covers high-risk systems: Annex III systems (automated recruitment, credit scoring, school admissions…) move to 2 December 2027, and Annex I systems to 2 August 2028. Article 50 transparency rules — the ones that cover chatbots — keep their original date of 2 August 2026.

What are the penalties for non-compliance with the transparency rules?+

Breaches of the AI Act’s transparency obligations can be fined up to €15 million or 3% of global annual turnover, whichever is higher. In France, the CNIL is preparing to act as market surveillance authority. For an SMB, the practical risk is an inspection revealing an unaudited chatbot script.

Is a mention in my terms of service enough to comply?+

No. The disclosure must be clear and appear at the start of the interaction — an explicit message such as “You’re chatting with an AI assistant.” A line buried in the terms of service or the page footer does not meet the requirement.

Free resource

The self-assessment grid: 20 tasks AI can automate

Sales, admin, support, operations: the 20 tasks AI agents already handle in SMEs — with, for each one, the tell-tale sign that your team is concerned.

Read next