The 7 Mandatory Clauses Updated for Late-2026 UK Enforcement
Blanket ChatGPT bans fail in practice—employees route prompts through personal phones while your ISO/IEC 42001 Statement of Applicability becomes inaccurate. Build a clear traffic-light framework incorporating four statutory triggers introduced or clarified across 2026:
- 1. ISO/IEC 42001 Clause 4.3/8.1 Scope & Shadow AI Ban: Restrict work prompts to approved enterprise workspaces with Zero Data Retention; register all SaaS AI plugins and Model Context Protocol (MCP) connectors.
- 2. ICO Agentic AI Ownership & PECR Regulation 6: Assign a named human owner and least-privilege API permissions to every autonomous agent, and trigger prior PECR consent whenever agents access/store terminal device session state.
- 3. UK GDPR Personal Data & Zero Retention: Prohibit entering customer PII, special category data, or proprietary code into unapproved models.
- 4. DUAA 2025 Articles 22A–22C Meaningful Human Oversight: Ensure human reviewers have authority, competence, and time to override AI recommendations, with appeal/contestability logging.
- 5. Equality Act 2010 Section 60 Recruitment Guardrail: Ban HR or recruitment chatbots from asking or recording pre-offer health/disability questions and preserve Section 20 human bypasses.
- 6. EU AI Act Article 4 Literacy & Article 50 Disclosures: Log staff AI training and watermark/disclose external generative AI interactions where outputs reach EU users.
- 7. 2-Hour Security & DPIA Intake Reporting: Route prompt data leaks immediately to the DPO under UK GDPR breach timelines.
Backing Policy with Runtime DLP & Red-Teaming
Manage staff policy acknowledgements inside Naq, ISMS.online, or Vanta, enforce browser-level PII redaction via Harmonic Security (Mindgard vs Lakera covers model red-teaming and sub-50ms production guardrails). Use the copyable template below to launch within 24 hours.
Copyable UK Generative & Shadow AI Policy Template
Copy or download this plain-English baseline template and adapt the bracketed fields to your organisation before circulation.
UK ACCEPTABLE USE OF GENERATIVE AI & SHADOW AI POLICY TEMPLATE (2026)
Organisation: [Company Name Ltd]
Policy Owner: [Data Protection Officer / CTO]
Review Cycle: Quarterly (Next Review: [Date])
1. PURPOSE & ISO/IEC 42001 SCOPE (CLAUSES 4.3 & 8.1)
This policy governs all external, embedded SaaS, and internal Generative AI or Agentic AI systems across all UK employees, contractors, and agency staff. Unlogged tools risk invalidating UKAS ISO/IEC 42001 scope (Clauses 4.3 & 8.1).
2. APPROVED TOOLS, SHADOW AI & NAMED AGENT OWNERS (ICO OCT 2026)
- Permitted: Only workspace-licensed enterprise AI accounts listed in the central AI Register (Zero Data Retention enabled; model training disabled).
- Prohibited ("Shadow AI"): Personal free-tier accounts, unapproved browser extensions, or unvetted Model Context Protocol (MCP) connectors.
- Agentic AI Ownership: Every autonomous AI agent must have a named human owner, hard-coded least-privilege tool boundaries, and PECR Regulation 6 terminal consent where session storage is used.
3. UK GDPR, DUAA 2025 & DATA HANDLING RULES
- NEVER paste personal data (names, emails, HR records, NI numbers, customer tickets), special category data (Article 22B), financial records, or proprietary source code into unapproved AI tools.
- Verify vendor DPAs (UK GDPR Art. 28) and living AI DPIAs (Art. 35(3)(a) + Legitimate Interests Assessment) prior to rollout.
4. HUMAN OVERSIGHT & CONTESTABILITY (DUAA 2025 ARTS. 22A–22C & EQUALITY ACT)
- Automated Decision-Making: Ensure named reviewers have technical competence, override authority, and time ("meaningful human involvement"). All solely automated significant decisions must embed Article 22C notice, representation, human review, and appeal logging.
- Employment & Recruitment: Never deploy chatbots to ask or log pre-offer health/disability questions (Equality Act 2010 Section 60), maintain Section 20 human-interview adjustments, and run disparate impact audits (Section 19 / Essop).
5. SECURITY, LITERACY, REPORTING & ENFORCEMENT
- Complete logged AI literacy training (EU AI Act Art. 4) and clear Art. 50 AI interaction disclosures where outputs reach EU users.
- Report accidental PII/IP prompt exposure to [[email protected]] within 2 hours, and route data protection complaints through [[email protected]] per DUAA 2025.