Starting August 2, 2026, your smart speaker in the EU is legally required to tell you it’s an AI. Ask Alexa for the weather, and before she checks the forecast, she has to announce she’s a machine. Same for Google Home. Same for Siri. This is not a hypothetical. It is Article 50 of the EU AI Act, and it takes effect in three weeks.
The rule is specific and leaves little room for creative interpretation. The disclosure must be audible and front-loaded — meaning it comes at the start of the interaction, not buried after five minutes of conversation. Even if your device already displays a bot icon, the use of natural language means it does not qualify for the narrow ‘obvious exception.’ Voice assistants using human-like speech must say so out loud, every time.
Tech companies face a fragmented reality. Article 50 transparency obligations are locked for August 2, but the broader EU AI Act timeline is in flux. The European Parliament voted to delay compliance for high-risk AI systems until December 2, 2027. The Council of the EU hasn’t approved that shift yet. Because Article 50 is legally separate from the high-risk requirements, it stays on the original 2026 schedule. Companies must prepare for an immediate transparency deadline while tracking a potential 16-month delay for their more complex AI infrastructure.
Amazon, Google, and Apple haven’t said a word about how they’ll comply. No public statements on what their voice assistants will actually say, how often they’ll say it, or whether they’ll try to minimize the interruption. The financial stakes are significant: penalties under the EU AI Act reach up to €15 million or 3 percent of global annual turnover per violation — not per day, as some reports have suggested, but per violation. For companies generating tens of billions in annual revenue, the fines are large enough to matter but small enough that the real risk may be the design friction.
For a device engineered to feel seamless and invisible, a mandatory verbal disclosure at the start of every interaction is a significant design challenge. The average annual compliance cost per AI system is estimated at approximately €29,277, but the real cost for these companies is the friction added to their products. A smart speaker that has to announce its nature before every answer is a fundamentally different user experience than one that just responds.
There is a narrow grace period for older hardware. Article 50(2) gives companies until December 2, 2026, to meet marking obligations for synthetic audio on systems placed on the market before August 2. But this does not exempt them from the core transparency requirement for new interactions. The regulation applies extraterritorially — any provider targeting the EU market must comply, regardless of where the company is based.
For consumers, the change is straightforward. The device that used to just answer your question will now tell you it’s a machine first. Whether that disclosure becomes a quick, unobtrusive greeting or a clunky interruption depends on implementation choices companies haven’t revealed yet. As August 2 approaches, you’ll be hearing more from your smart speaker — whether you asked for it or not.