Show Notes
Ofcom fined Tismi four hundred and thirty-seven thousand five hundred pounds after about three hundred and sixty-six thousand sub-allocated phone numbers were exposed to potential misuse for over two years, with evidence of nuisance and scam calls tied to the numbers. 📞⚠️
The episode also covers ElevenLabs’ Eleven Realtime launch, which it says turns speech-to-speech into a complete voice agent, and Mistral’s Voxtral 3 voice-agent kit, priced at four US cents a minute in one configuration.
The practical takeaway is that phone-number onboarding, ongoing monitoring and misuse reporting are now part of the risk picture for any voice deployment, while the model launches still leave key caveats open: pricing, what is included, and how performance holds up on real calls.
Next, attention turns to the fuller Tismi decision, the new Xtra Phones UK investigation, ElevenLabs’ Realtime pricing and whether Mistral clarifies its per-minute voice-agent costs. Signal and Story examines what checks telecoms providers should make, what disappears when voice systems are assembled as one model, and what the published price actually covers.
More from Signal and Story
The FCA published examples of good and poor AI practice in financial services, including chatbots limited to approved sources, escalation routes and customer choice, alongside failures that left people unable to get proper help. 🤖📞 The episode also covers NatWest’s adapted speech recognition for UK accents and banking language, the CMA’s proposed strategic market status for Microsoft’s business software ecosystem, and SpaceXAI’s release of Grok Voice 3.0 at eight US cents a minute. The practical takeaway is that AI systems need testing against real accents, real customer journeys and real vendor dependency risks, not just benchmark claims. The episode also flags that the FCA’s examples are illustrative rather than new rules, and that the Grok pricing and leaderboard figures are snapshots, not a live production bill. What to watch next is the FCA’s direction on significant AI failures, the outcome of the CMA consultation on Microsoft and Copilot, and whether voice-AI buyers can balance price, latency and stability against the models they actually need. Signal and Story examines how customer service automation, platform choice and speech recognition shape enterprise AI operations.
The ICO opened a six-week call for evidence on agentic AI, asking how systems that can use tools and act with limited human oversight are being governed, and stressing that autonomy is not an excuse for poor compliance 🤖📣 The episode also covers complaints about automated nuisance calls rising from 1,374 in July to 3,684 in September, plus the Royal Mail Christmas advert row after a family said the AI-generated voice sounded like their late father. For operators, the message is to pin down what a system can do, what data it can reach and who remains accountable. The episode flags open questions on conversational voice agents, because PECR clearly covers recorded-message automated calls but not necessarily live AI dialogue. Next up: the ICO’s evidence call runs to 20 November, the PECR review may clarify live conversational calls, and the Royal Mail dispute leaves the voice industry facing tougher resemblance checks. Signal and Story examines how regulators define agentic AI, how automation changes call consent, and how synthetic voices should be governed.
BT Business launched AI Receptionist for UK firms, an always-on phone agent that handles enquiries, bookings and confirmations, can integrate with Outlook and Google Calendar, and is sold in monthly rolling bundles of inbound minutes. 📞🤖 The episode also covers ElevenLabs’ three-hundred-million-dollar employee tender that valued the company at twenty-two billion dollars, plus Microsoft’s release of MAI-Transcribe-2-Streaming and MAI-Voice-2.1 models with clear per-hour and per-character pricing. The practical takeaway is that voice automation is becoming easier to buy and meter, but the caveats matter: BT does not publish bundle prices, ElevenLabs’ performance claims are company-reported, and Microsoft marks its streaming transcription model as public preview and not recommended for production. Watch for BT’s real-world take-up and any independent view of call-handling quality, Microsoft’s move from preview to an SLA-backed service, and whether that fifty-four-cent introductory price changes after the end of the year. Signal and Story examines what AI phone agents cost, how quickly they work, and where production readiness still falls short.