"The AI said it, not us" has been tried as a defence, and it failed. Six questions that should have an answer before your voice agent speaks about price, dates or coverage.
The trade press spent the spring reporting that analysing a caller's tone of voice would be reclassified as high-risk on 2 August. That deadline moved to December 2027. The prohibition that actually applies reached the contact centre in February 2025 — and it is about your staff, not your customers.
Most companies pick the voice their agent speaks with in under a minute, from a list. That choice is turning into a purchase – one with an origin, a contract and a marking obligation attached – and the three questions that make it safe cost nothing to ask now.
Norway's telecom regulator has ruled that call centres may not use 8-digit mobile numbers as their caller ID. If your voice agent makes outbound calls there, the number is a compliance question before it is a marketing choice.
Last week's attack on Wall Street's largest hedge funds did not come through the payment flow. It came through the service desk, in a voice the employees recognised.
Sentiment analysis, tone-of-voice detection and frustration alerts ship with most modern telephony platforms. That kind of feature is classified as high-risk under the EU AI Act, but the obligations were pushed back to 2 December 2027 – and pointed at your own staff it has been prohibited since February 2025.
Since 2 August 2026, AI systems that interact directly with people must disclose that they are AI. For anyone running a voice agent on the phone, that is no longer an ethical judgement call – it is an obligation with fines attached.
Most Voice AI projects do not fail due to intelligence. They fail due to accountability.