TL;DR

The UK’s AI Security Institute (AISI) has announced a research partnership with ElevenLabs, a London-based voice AI company, to investigate safety and security in voice AI systems. The collaboration will focus on whether people can tell AI from humans in real-time conversations and how perceived identity affects user trust and behaviour.

What the partnership covers

The research will examine two core questions that are becoming more pressing as voice AI enters mainstream use. First, how well can people actually distinguish between AI and human speakers during live voice conversations? Second, how does the perceived identity of a conversational partner — whether someone believes they are talking to a person or a machine — shape their trust and behaviour?

These are not abstract concerns. Voice AI is being deployed in customer service, healthcare triage, sales and personal assistance. If users cannot reliably tell whether they are speaking to an AI, the implications for informed consent, manipulation and trust are significant.

How the collaboration works

The agreement enables two-way collaboration between AISI and ElevenLabs, covering shared research, investigation of emerging risks specific to voice AI, development of mitigation strategies and best practice guidance. ElevenLabs, which has become one of the leading companies in AI voice synthesis and cloning, published its own announcement confirming the partnership.

AISI said that developing evidence about “the impact of real-time AI voice conversations on humans” is necessary for deploying these systems “responsibly and securely.”

Looking forward

The partnership reflects growing regulatory attention to voice AI specifically, rather than AI as a broad category. As voice becomes a primary interface for AI interaction — through smart speakers, phone systems and wearable devices — establishing safety baselines for how these systems affect human behaviour and decision-making will be an important piece of the UK’s wider AI safety agenda.