Version 1.0 · Issued 24 May 2026 · Reviewed monthly
Nure Group internal AI register
Nure Group Limited is itself a deployer of multiple AI systems. This register satisfies the implicit precondition of Article 4 (a deployer must know what AI its staff use). We publish it because we expect our customers to do the same.
| Tool | Use case | Risk | Status |
|---|---|---|---|
Claude (Sonnet / Opus / Haiku) Anthropic · LLM chat / agent | Product development, code review, draft authoring, internal Q&A | limited | approved |
ChatGPT Team OpenAI · LLM chat | Occasional drafting, brainstorming | limited | approved |
GitHub Copilot (Business) GitHub / Microsoft · Coding assistant | Code completion in IDE | limited | approved |
Microsoft Copilot in Microsoft 365 Microsoft · Productivity AI | Drafting, summaries inside Word / Outlook | limited | pending review |
Anthropic Claude (in-product NureComp AI features) Anthropic · In-product LLM (customer-facing surface) | Powers NureComp smart-assignment, NL query, Trainer, AUP draft for Pro/Business customers | limited | approved |
Recruitment ATS AI Various · High-risk AI (Annex III #4) | Not in use — flagged for future review when team hiring begins | high | blocked |
Controls applied (per tool)
- Claude (Sonnet / Opus / Haiku) — No-training-on-inputs contract; EU SCCs in place; sensitive prompts use redacted sample data; conversations retained 30 days per Anthropic policy.
- ChatGPT Team — No-training-on-inputs (Team tier); EU SCCs in place; no client / staff PII pasted.
- GitHub Copilot (Business) — Business-tier no-training-on-customer-code contract; duplicate-detection filter enabled; no production secrets or PII in scope.
- Microsoft Copilot in Microsoft 365 — M365 tenant data-boundary applies; awaiting SharePoint permissions audit before broader enablement.
- Anthropic Claude (in-product NureComp AI features) — Per-org spend cap; per-feature opt-out; system prompts include legal/compliance guardrails; every call logged in aiUsageEvents; mandatory "not legal advice" disclaimer on every Trainer response.
- Recruitment ATS AI — Blocked pending DPIA + Equality Act 2010 bias audit before adoption. Founder commits not to deploy any ATS AI until both are completed and recorded.
Review schedule
Reviewed monthly by the founder. Any new tool adopted requires entry in this register before first operational use. New Annex III high-risk tool requires a DPIA before adoption regardless of urgency.
Published at /internal/ai-register · Linked from Trust Centre · Source of truth for the AI AUP self-acknowledgment at /internal/ai-aup.