For years the EU AI Act was the thing everyone name-dropped in meetings and nobody had actually been punished under. That changed on 2 August. The European Commission’s AI Office, working alongside national authorities, can now enforce the law, and a fresh set of transparency rules switched on the very same day.
The rules that reach your desk
Three obligations matter even if you never read a single legal clause. Chatbots and other interactive systems now have to tell people they are talking to a machine, not a patient human on a headset. Deepfakes, meaning any image, video or audio made or edited with AI, have to be labelled. And AI-generated or altered content needs machine-readable marks built in, so other software can flag it without a human squinting at it.
The Commission’s line is that this cuts down on deception and manipulation and gives people a fair shot at knowing what is real. For businesses it also means a clear checklist instead of a vague vibe about ‘responsible AI’.
Fines that focus the mind
The AI Act’s penalty ceiling for breaching these duties runs to 15 million euro or 3% of worldwide annual turnover, whichever is higher. For a corner shop that is theoretical. For a company the size of Meta or Google, 3% of global turnover is not a rounding error, it is a genuinely rotten quarter.
Who is already under the microscope
CNBC reported that Anthropic and OpenAI are among the frontier labs facing fresh scrutiny under the new enforcement powers. The Commission also published a first list of more than 180 organisations that have signed its Code of Practice on transparency of AI-generated content, which is the voluntary way of showing regulators you are trying. Signing is not mandatory, but it is a lot cheaper than the alternative.
What to actually do this week
If you run a chatbot on your site, add a plain line telling visitors it is AI. If you publish AI images or video, label them and keep whatever provenance marks your tools offer switched on. If you work in marketing or anything creative, assume disclosure is now the default rather than a nice-to-have. None of this is hard. The expensive part is ignoring it until a complaint lands, and yes, the Commission has built both a complaints tool and a whistleblower channel for exactly that.
The bigger picture is that Europe and the United States have wandered off in opposite directions. The EU has gone risk-based and enforceable. Washington has leaned towards lighter-touch voluntary standards. If you sell into Europe, the EU version is the one that can send you an invoice.
Did you know: the AI Act was first proposed back in April 2021, which in AI-industry time is roughly three geological eras and about forty model releases ago.
Sources
- European Commission: Commission starts enforcing AI Act rules and new transparency requirements on 2 August
- CNBC: Anthropic, OpenAI among firms facing new scrutiny under EU AI Act enforcement powers
- Help Net Security: EU begins enforcing AI Act
Related on Top Tool Stack: ChatGPT Just Handed Free Users Unlimited Chats and a Think Button · ChatGPT Can Now Book Your Dinner Table, With a Few Catches