Approved AI tools

Your staff use AI tools: chat assistants, code assistants, office copilots, AI features inside CAD and ERP. The AI-use policy says in one sentence that only approved tools may be used and that no confidential data goes into public ones. The approved-AI-tools register turns that sentence into a maintained list: tool, vendor, your decision, allowed data classes, conditions, owner and next review. The policy references it, the employee portal shows it, the Article 4 evidence pack contains it. The register is an Enterprise plan feature.

What the register is — and is not

• A measure you evidence. The AI Act prescribes neither a list of approved tools nor data classes. Article 4 asks deployers for measures to ensure a sufficient level of AI literacy; a maintained, communicated list of permitted tools and permitted data is a reasonable measure and evidence of it — nothing more. Resilic never says the register is required by law.
• What the data-class rule serves: your data protection (GDPR Art. 32 security of processing, Art. 28 processors), your trade secrets (the German trade-secret act protects only what reasonable measures protect), NIS2 Art. 21 where you are in scope, and your own information security. Which of these apply to you is not something Resilic determines.
• No monitoring. The register holds decisions, not usage. There is no agent, no proxy, no content inspection and no per-employee usage data. A published list triggers no co-determination; monitoring would. The register stays deliberately on this side.
• You decide. Resilic suggests nothing beyond a starter list of common tools; every decision, data class and condition is yours, recorded with name and date.

Recording a tool

Under Comply → AI tools. Add common tools seeds a starter list of widely used tools (ChatGPT, Microsoft 365 Copilot, GitHub Copilot, Claude, DeepL, …) as Under review — with name, vendor, kind and website, without any decision. The action can be repeated; names already present are skipped. Your own tools go in through New tool.

Per tool you record the facts: name, vendor, kind (chat assistant, code assistant, office copilot, embedded feature, API, other), website, the processing location per the vendor (EU, outside the EU, unknown), whether a data processing agreement is in place and whether the vendor commits not to train on inputs, and who in your organisation owns it. The vendor facts are what you read in the vendor’s terms — Resilic does not verify them.

The decision

The decision is a separate step, recorded with the signed-in member and the time:

• Under review — the starting state. Staff do not see the tool.
• Approved — with at least one allowed data class.
• Restricted — with data classes and conditions, for example “company account only; no customer data”.
• Prohibited — without data classes; a reason can be left in the conditions.

The data classes are Public, Internal, Confidential, Personal data and Customer data. The next review defaults to one year after the decision; due reviews are flagged in the register.

Policy, employee portal, evidence pack

• The Acceptable use section of the generated AI-use policy references the register and says it is maintained separately and may change without a new policy version. A change in the register therefore invalidates no attestation — the list stays current without everyone re-attesting.
• Staff see the decided tools with data classes and conditions on their attestation page; tools under review do not appear there.
• The Article 4 evidence pack contains an “Approved AI tools” section with every decision, data class, condition, decider and review date.

Applicability note

The register is guidance and evidence, not legal advice. Whether and which duties under GDPR, the trade-secret act or NIS2 apply to a tool is for your own data-protection and security process to determine.