4 September 2026

AI news of the week: the AI Act postpones deadlines, WhatsApp starts charging and three AIs go down on the same day

Every week I go through the AI news that really affects a small business or a self-employed person. No hype: what happened, and what you need to do.

The AI Act postpones high-risk obligations to 2027 (but not transparency)

Regulation (EU) 2026/1744 of 8 July 2026 — the so-called Digital Omnibus on AI — is now in the Official Journal of the EU and directly applicable in all member states. It postpones the obligations for Annex III high-risk AI systems to 2 December 2027, and those for systems embedded in regulated products under Annex I to 2 August 2028. What is not postponed are the transparency obligations of Article 50, enforceable since 2 August 2026.

How does this affect me?

As a person: when you talk to a customer-service chat, they have to tell you it is a machine. And AI-generated content — images, audio, video — has to be identified as such. If you see a company not doing it, it is no longer a matter of style: it is an obligation.

If you run a business or are self-employed: check two specific things this week. First, the welcome message of your automated chat, answering service or assistant: it must make clear that an AI system is replying. Second, the images or videos you post on social media: if you generated them with AI, say so. The training obligation of Article 4 remains in place since February 2025 (I covered it here). And if someone calls offering you an "official AI Act certificate", hang up: that accreditation does not exist.

WhatsApp will stop being free for replying to customers from 1 October

Until now, if a customer wrote to you on WhatsApp and you replied within the following 24 hours, those messages generated no charge from Meta. From 1 October 2026, Meta will charge per message for service messages sent within that window, and also for the utility templates that until now were included. The new rates were published on 1 September. Note: this affects the WhatsApp Business Platform (the API), the one you use if a provider automates your WhatsApp. The WhatsApp Business mobile app remains free.

There is a detail that costs money and almost nobody explains: Meta warns that anyone without a payment method registered before 30 September will no longer be able to send service messages from 1 October. It is not a fine; the channel simply goes silent.

How does this affect me?

If you run a business or are self-employed: if your WhatsApp is connected to a CRM, a shop or a bot, do two things this week. First, ask your provider whether the WhatsApp account has an active payment method; if they manage it, get it confirmed in writing before 30 September. Second, ask how many service messages you sent last month and what they would have cost at the new rates. With that number you decide. Often the conclusion is not to leave WhatsApp, but to close conversations faster and stop using ten messages for something that needs three. If you do not use the API and reply by hand from your phone, you do not need to do anything.

ChatGPT, Claude and Grok went down almost simultaneously

On the morning of 3 September, three of the most widely used AI assistants stopped working almost at the same time. The services of OpenAI (ChatGPT), Anthropic (Claude) and xAI (Grok) were inaccessible for part of the morning, and service was gradually restored over midday. None of the three companies has confirmed a common cause.

How does this affect me?

As a person: if you use AI for something with a deadline — an application, a procedure, a document — do not leave it to the last day. It is not a tool you can take for granted at a specific hour.

If you run a business or are self-employed: ask yourself one question and write the answer on a sheet of paper: if tomorrow morning the tool that answers your customers or processes your invoices goes down, who does it? If the answer is "I don't know", you have this week's job. The practical rule is simple: no process that can lose you money or customers should depend on a single provider without a documented manual alternative. You do not need to hire two; you need to know what to do while one is down.

Phishing now arrives by QR, by WhatsApp and with a familiar voice

During August there was a surge in fraud campaigns combining fake QR codes, WhatsApp and Telegram messages imitating well-known platforms, and AI-cloned voices impersonating company managers to request a payment or a change of bank account. The recommendation is always the same: verify through a different channel from the one the request arrived on.

How does this affect me?

As a person: be wary of QR codes stuck over official signs — on parking meters, in restaurants or hotels — and make payments from the official app or website, not from whatever the QR opens.

If you run a business or are self-employed: put in place a one-line rule and tell the whole team: no change of account number and no urgent payment is validated through the same channel the request arrived on. If someone calls with the manager's voice asking for a transfer, hang up and call back on the usual number. A familiar voice is no longer proof of anything. And if you have people on call or a small team, that is when having it written down matters most.

Sources

Meta's documentation on the WhatsApp pricing changes; the official text of Regulation (EU) 2026/1744 on EUR-Lex; a report on the 3 September outage (in Spanish); and the security alerts of INCIBE, Spain's cybersecurity institute.

To keep track of who has written to you and who is waiting for a reply before these WhatsApp changes catch you off guard, you need a contact follow-up tool; MemorIA, which I built, is one example of how it can be approached.