A UK briefing maps psychological risk across AI assistants

The design

On 4 February 2025, the UK's Ada Lovelace Institute published 'Delegation Nation: Advanced AI Assistants and why they matter,' a policy briefing within its 'How can (A)I help?' research project. The briefing addresses 'Advanced AI Assistants' broadly – voice assistants, agentic tools and companion-style chatbots together – rather than companion products alone, and it is a literature synthesis rather than a newly commissioned survey; readers expecting a dedicated companion-AI study should note this scope.

What the evidence says

The briefing argues that systems capable of 'exert[ing] considerable influence over users' mental and emotional states' raise underexamined policy questions, and in support it cites named peer-reviewed research rather than presenting new data of its own. Among its citations is Linnea Laestadius and colleagues' 'Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika,' published in New Media & Society in October 2024 – a title that itself states the study's method (grounded theory analysis) and subject (Replika). The Institute's briefing also cites William Seymour and Max Van Kleek's CSCW paper on trust and anthropomorphism in voice assistants. Ada Lovelace Institute is characterising and citing this research, not conducting it.

What it asks of people

The briefing frames advanced AI assistants, companion products included, as raising 'especially difficult questions about... the psychological impact of people's relationship to AI,' asking policymakers and users alike to weigh convenience and companionship against ceding influence over 'user thought and behaviour' to a small number of developers, in the briefing's own phrase.

Privacy and safeguards

As a policy briefing rather than a regulator's order, the document recommends areas needing scrutiny instead of mandating a remedy; its own bibliography shows it works primarily by pointing to existing peer-reviewed and legal research rather than by auditing any single company's current practice. Its analysis is current as retrieved on 16 September 2026 and may be superseded by the Institute's later, longer report 'The Dilemmas of Delegation,' published in November 2025.

• Does grouping companion chatbots with task-focused AI assistants under one policy label obscure risks specific to relational products?
• What would the Laestadius grounded-theory findings add if read directly rather than through a policy briefing's citation?
• Which of the briefing's cited concerns have since been addressed by named regulation, and which remain open?

This is credible secondary synthesis rather than primary evidence about any single companion product; its value lies in mapping where the underlying peer-reviewed research already sits, not in adding to it.