Baroness Campbell of Surbiton delivered a speech in the House of Lords on October 13, 2026 in an AI recreation of her own voice, which ElevenLabs and AFP both describe as the first time a peer or MP has used an AI-generated voice to address either house of the UK Parliament. The speech was during the Second Reading of the Health Bill. The words were hers; the voice was a clone built from old recordings of how she sounded before spinal muscular atrophy weakened it.
This is a rare AI story where the use case is unambiguous: an assistive technology, approved in advance by the institution, used by the person whose voice it is.
TL;DR: the facts
| Question | Answer |
|---|---|
| Who? | Jane Campbell, Baroness Campbell of Surbiton, 67, an independent crossbench peer since 2007 |
| What? | Delivered a speech with an AI recreation of her voice |
| Where and when? | House of Lords, Second Reading of the Health Bill, October 13, 2026 |
| Tech provider | ElevenLabs |
| Who approved? | The House of Lords Procedure Committee |
| Who wrote the speech? | Baroness Campbell |
| Topic | Personal Health Budgets (PHBs) |
| Why now? | Her voice has gradually deteriorated; since 2012 she had relied on an assistant to read her speeches |
How did the AI voice work?
Text-to-speech illustration for Baroness Campbell's AI voice speech
According to ElevenLabs, Baroness Campbell found its voice cloning technology on her own and uploaded old clips of her speaking to create a voice that replicates the tone, rhythm and pitch of her original voice. She then used the company's text-to-speech generator to turn her written speech into audio, which was played through an assistive device in the chamber.
AFP adds that the company "worked with her briefly on her voice clone so that we could fine-tune it and make sure that it was exactly how she wanted," in the words of ElevenLabs' Gabi Leibowitz, a speech therapist by training. Leibowitz said 30 seconds of audio can be enough to create a clone, but 30 minutes to an hour gives the best quality. AFP also reports that the technology allows what ElevenLabs calls "emotion tags" that control the tone of the voice. Leibowitz said that before generative AI, someone would have to record almost 400 sentences before losing their voice, and the result was "relatively robotic" and limited to one language.
ElevenLabs says the speech carried an emotional range she had not been able to deliver in decades. Our coverage of the company's newest speech models, Eleven v4 and v4 Turbo, explains the expressive controls behind this kind of delivery. ElevenLabs' own page for the announcement also advertises Eleven v4 as its "most emotive model yet", though the post does not say which model version was used for this speech.
What did she say?
Introducing the technology to the chamber, the 67-year-old said (per AFP): "This my lords is cutting-edge technology and it will enable me to contribute to debates in my own, once-clear voice before it began to deteriorate due to my disability." She added: "So my lords, may I introduce you to Baroness Campbell Mark 2."
In ElevenLabs' account she said: "I used AI to speak my own words, not to put words into my mouth. The speech, the arguments and the views are entirely mine. It's my voice, my style, my words. A totally empowering experience."
The speech itself argued that Personal Health Budgets, funds allocated to individuals in England for treatments, equipment and personal care, must stay accessible to the disabled people who depend on them and be extended to others who could benefit. "I depend upon a Personal Health Budget myself," she said.
What did Parliament approve, and why does the history matter?
The Procedure Committee, which considers Lords protocol, granted permission, per AFP citing the Lords press office. ElevenLabs notes that in 2012 the same committee allowed her to bring her assistant into the chamber to read for her, a departure from a 1707 rule that prohibited non-peers from crossing the bar and speaking in the House. This AI-voiced speech is the second accommodation of its kind for her.
Lord Speaker Lord Forsyth of Drumlean said: "While there are justifiable concerns about some uses of AI, when used in this and many other ways, for good, we must recognize its capabilities and transformative potential."
The institutional sequence is the notable part. Rather than banning or ignoring the technology, an old chamber with eleventh-century origins ran a request through its normal procedure committee and approved a specific, bounded use. That is a template other legislatures and courts could follow.
Why this is different from the voice-clone stories you usually read
Most AI voice news is about fraud, deepfakes or unauthorized likeness. We covered a Chinese top court's ruling on AI clones and personality rights, and a steady stream of speech models like Fish Audio Drama 3 and the local, open-source VoiceStudio that make cloning cheap. This case sits at the other end of the spectrum. The elements that make it uncontroversial are also a checklist for any similar deployment:
- Consent. The voice belongs to the speaker and she chose to clone it.
- Authorship. The words were written by the speaker, not generated for her.
- Disclosure. She told the chamber what the technology was before it spoke.
- Prior approval. The institution's rule-making body approved it.
- A clear purpose. Accessibility, not persuasion or impersonation.
Remove any of these and the same technology looks very different. A synthetic voice reading generated words without disclosure in a political setting would raise the harms critics worry about.
What is the wider assistive-voice context?
ElevenLabs says its technology has been used to restore the voices of more than 11,000 people who lost their voices through injury or illness, and that it has committed to helping one million people with voice loss. Those are company figures. It points people to its Impact Program page. The company's business mostly runs in other directions, such as AI receptionists for small businesses and enterprise conversational agents.
Competition in synthetic speech is intense, with Google's Gemini 3.8 Flash TTS among the rival offerings, which matters for accessibility because lower costs and local models make personal voice banking more reachable. People who expect to lose their voice, for example from motor neurone disease or throat surgery, can bank recordings in advance; the quality bar for old recordings is also much lower than it used to be.
How would someone do this responsibly?
Based on what ElevenLabs and AFP describe, a sensible process for anyone considering a personal voice clone looks like this:
- Gather source audio you own. Old home videos, voicemails or recordings of talks work. More audio gives a better clone: the company's guidance is 30 minutes to an hour, though 30 seconds can work.
- Clone only your own voice, or one you have written permission to clone. Reputable services ask for consent verification for exactly this reason.
- Tune it to feel like you. The fine-tuning step Leibowitz describes, checking the clone is how the person wants to sound, mattered here because the goal was authenticity, not just intelligibility.
- Script and tag the delivery. Emotion tags and pacing controls let you shape emphasis. Listen back before any live use.
- Plan the playback. In the Lords the audio was played through an assistive device, so the rehearsal needs to cover the hardware chain, not only the model.
- Disclose and get permission. Tell the audience, and ask the venue's rule-makers first, as she did.
None of this removes the risks of voice cloning in general, and the tools that make accessibility cheaper also make impersonation cheaper. If you run AI agents that can speak or act on a person's behalf, you need controls beyond consent screens; AgentBeam, from the explainx.ai team, is an agent security platform that stops AI agents before they take dangerous actions.
What are people asking?
Is the voice really hers? It is a clone built from her old recordings, tuned with her input to sound as she wanted. It is her voice recreated, not her live voice.
Did AI write the speech? The reporting says no; she wrote it. ElevenLabs quotes her explicitly on this.
Could this be abused in Parliament? The approval process is the control. Nothing in the sources we read describes a general rule change allowing AI voices; the permission described is specific to her.
Is this available to anyone? The consumer tools are, but the chamber permission is not general. Any other member would presumably need to go through the same committee, though we have not seen guidance on that.
What to take from it
For builders working on voice products, this is a useful reference case: design for consent, disclosure and the user's own authorship, and make the output controllable enough (tone, pacing, emotion) that the result sounds like the person, not like a generic narrator. For institutions, it shows an approval path that is measured and specific. And for readers who think of AI voice only as a fraud risk, it is a reminder that restoring a person's ability to take part is a major use of the same technology.
What we could not verify: the exact ElevenLabs model version used, and whether the Lords will publish guidance for other members. Both ElevenLabs and the Lords press office are the best places to check.
Related reading
- ElevenLabs Eleven v4 and v4 Turbo voice models
- China's top court on AI clones and personality rights
- Fish Audio Drama 3 open weights
- VoiceStudio: an open-source local ElevenLabs alternative
- ElevenLabs Reception for small businesses
- Gemini 3.8 Flash TTS and voice design
Details reflect ElevenLabs and AFP reporting on October 13, 2026.