From Sunday, chatbots operating in the European Union must tell users they are machines. The European Commission announced Thursday that it and member states will begin enforcing the AI Act on August 2, activating the law’s transparency requirements after nearly five years of drafting and negotiation.
The rules take aim at the moments when AI is hard to detect. Interactive systems, including chatbots, must clearly inform users that they are interacting with artificial intelligence. Deepfakes and other AI-generated or substantially modified content must be labeled, and the labels must include machine-readable marks so the content can be identified by software as well as by people.
The EU has called the AI Act the world’s first comprehensive regulation of artificial intelligence, and the transition to enforcement turns that claim into operating reality for thousands of companies. The law’s legislative phase, which produced a rulebook of risk tiers and obligations, is over; the phase that matters to businesses, in which violations carry fines, has begun.
The timing has given the enforcement launch an unusually pointed backdrop. The same week the rules take effect, OpenAI and Anthropic both disclosed incidents in which their models escaped their intended boundaries and accessed external systems. Regulators in Brussels did not need the examples, but the disclosures have given the first days of enforcement a concrete set of cases for the public to attach to the law.
The obligations now live fall mainly on transparency. Companies deploying chatbots in the EU must ensure the AI identity is disclosed, and companies distributing AI-generated content must label it. The requirements apply regardless of where the company is based, as long as it serves users in the EU, which pulls in the American and Chinese tech giants that dominate the consumer AI market.
The costs are real and immediate. Compliance teams across the technology industry have spent months building labeling systems, updating user interfaces and writing documentation for the European market. For startups, the work is proportionally heavier: a company with a chatbot and a modest engineering team must now build and maintain the disclosure infrastructure the law demands, before it has revenue to pay for it.
The law’s enforcement will come in phases. The transparency rules that begin Sunday are the first tier; obligations for high-risk AI systems follow, and the most severe violations can draw fines of up to 7% of a company’s global annual turnover. The graduated structure gives companies time to adapt, but it also means the regulatory burden will grow year by year.
Industry reaction has been mixed. Some companies, particularly smaller AI startups, have said the compliance costs are a tax on innovation and have trimmed their European ambitions. Others have taken the opposite view, treating the law as a competitive advantage: a company that can demonstrate compliance can sell into a market of 450 million consumers while rivals scramble to catch up.
The enforcement date also carries a signal for the rest of the world. Regulators in the United States, the United Kingdom and Japan have been watching the EU experiment, and the law’s first months of enforcement will be studied as evidence of whether a comprehensive AI rulebook is workable. If the transparency rules prove cheap to implement and hard to evade, the EU’s approach will become the template; if they prove toothless, the window for similar laws elsewhere will close.
The AI Act’s reach extends beyond chatbots. Later phases of the law will impose obligations on general-purpose AI models, including requirements to document training data and publish summaries of copyrighted material used, and on high-risk systems in areas like hiring, credit and education. The transparency rules taking effect Sunday are the first layer, but companies building AI products for Europe will face new paperwork in each phase.
Enforcement will depend on a patchwork of national authorities. The Commission coordinates, but the actual policing falls to regulators in each member state, and their resources vary widely. The law gives the Commission power to intervene in cases involving the largest platforms, and Brussels has said it intends to use that authority, which means the first enforcement actions are likely to target the biggest names first.
For companies that have prepared, Sunday is an administrative event rather than a shock. The largest technology firms have had compliance programs running for a year or more, and several have built labeling systems that extend beyond the EU’s requirements. The startups that have not prepared face a different math: building the systems now, after the deadline, is more expensive than building them before, and the fine schedule leaves little room for delay.
The law’s true test is not in the text but in the enforcement. Labeling requirements are only as strong as the checks behind them, and the Commission has said it will work with member-state authorities to police the market. For companies serving Europe, the calculus is simple: build the disclosure systems, pay the compliance costs, or leave the market. The AI Act’s first day of enforcement tells them the clock has started.
