AI transparency rules and where the UK is heading
Agentric Team ·
Where is UK AI regulation actually going? If you're deploying an AI receptionist, here's the honest map — what's binding now, what's coming, and why "disclose the AI" is the safe bet either way.
What's actually in force today
The UK doesn't (yet) have a dedicated AI Act — the approach so far is existing regulators applying existing law to AI:
- UK GDPR + ICO guidance — transparency about automated processing, fairness, data minimisation. The ICO's AI guidance expects people to know when AI is involved in decisions/interactions affecting them.
- Consumer protection law — misleading practices prohibited; presenting an AI as human arguably qualifies.
- PECR — covers the comms channels (marketing calls/messages) rather than the AI itself.
The government's stated approach: principles-based, regulator-led, sector-specific rather than a single AI law. Transparency, safety, fairness and accountability are the named principles — and disclosure sits squarely inside transparency.
The EU AI Act shadow
Even UK-only businesses should note the EU AI Act, which does explicitly require conversational AI to disclose itself. UK regulation generally converges in this direction, and building to the EU standard is the conservative posture regardless.
What's likely to come
Reasonable expectations for UK AI rules over the next few years:
- Formal transparency requirements for conversational AI (following the EU model).
- Duties around AI accuracy claims and oversight.
- Sector regulators (FCA, SRA, CQC) writing AI-specific rules for their domains — which will matter for professional services users.
What this means for a small business today
The durable practices that will survive any foreseeable regulation:
- Disclose the AI — upfront, always.
- Keep humans reachable — real escalation paths, not bot walls.
- Be transparent about data — recording notices, retention, privacy notices.
- Document the reasoning — lawful bases, supplier due diligence (the DPA).
- Stay in your lane — the assistant answers enquiries and books; it doesn't give regulated advice.
Agentric is built to that standard now — disclosure, boundaries, auditability. Whatever the regulation ends up looking like formally, those are the behaviours it'll require.