The regulation of delegation: Are AI advisers, agents and companions regulated in the UK?

Key Findings

• — No legislation specifically targets harms from AI assistants; coverage comes from horizontal instruments such as data protection and consumer protection, plus sectoral rules in financial services and legal professional regulation.
• — Protection applies on an all-or-nothing basis gated by threshold conditions, for example whether a lawyer-client relationship exists, so an assistant giving the same advice outside that relationship falls outside the rule.
• — Redress is impaired by the absence of transparency into how an assistant reached a decision and by the lack of settled legal standards for developers and deployers.
• — Individual redress places the burden on the user to notice a problem, while assistants are designed to encourage disengagement from scrutiny.
• — Nuanced, diffuse and social harms, including emotional-wellbeing harms and opinion-influencing, are poorly covered.
• — The introduction names Woebot, Nomi and Replika, and cites cases of assistants encouraging suicide and enabling delusional thought patterns.

Methodology Notes

Doctrinal legal analysis of England and Wales only, with no empirical component; published December 2025, so it predates subsequent UK regulatory movement. The underlying legal analysis was carried out by AWO and the briefing is written by Ada Lovelace Institute staff. The publication page gives 1 December 2025 in its structured metadata; the page was fetched with curl (WebFetch 403s this domain).

Sources