PHINEO.Workshop 2 · Smart Citizenship

AI at the CC desk

A field guide for using AI in your real work — safely, with every decision and the accountability staying with you. Inward on your own work · outward to nonprofits.

AI advises — you decide and you sign.
Draft & scan
Partner research
Human-in-loop
Application scoring
Draft & scan
CSRD / impact draft
Draft & scan
Stakeholder comms
You decide
Grantee / funding
Where AI fits your CC value chain. draft & scan human-in-the-loop you decide — never delegateCC owns the responsible-AI call.

1 Which tool?

Have Microsoft 365Copilot — safest, inside your data boundary
Google WorkspaceGemini — inside your boundary
Strongest all-rounderChatGPT / Claude — Team / Enterprise tier only
Legal blocks everythingLangdock (DE) / Mistral (FR) — EU-hosted, GDPR-capable*
TranslationDeepL (DE) — the easy yes
ESG / CSRDBriink, or SAP Joule for SAP shops

*GDPR-capable ≠ compliant out of the box: still needs a signed DPA + DPIA, and staff tools need Betriebsrat sign-off. Never a free consumer app for confidential data.

2 Do & Don't

Do

  • Use the approved tool
  • Treat every output as a draft
  • Give it role + audience + format
  • Own the decision

Don't

  • Paste confidential / partner data into a free app
  • Trust names, numbers, legal claims unchecked
  • Send AI text to a stakeholder unread
  • Outsource judgment to the model

3 Before & after — your guardrails

Pre-flight before you type

  • Approved tool for this kind of data?
  • Stripped anything confidential / personal?
  • Gave it the role, audience & format?
  • Pasted the source material it needs?

Post-flight before you use it

  • Facts, names, numbers, legal claims verified?
  • Any invented org / stat / source?
  • Sounds like us? Nothing internal leaked?
  • Would I put my name on it? → send.
PHINEO.Commands · Pitch · Point it outward

4 8 commands to steal

1Summarize a partner"Annual report → mission, who they serve, top 3 needs, red flags, 3 questions before partnering."
2Score applications"5 applications + our 4 criteria → rank, score each 1–5, flag the closest call." You still decide.
3CSRD / ESG draft"Notes → 3 CSRD impact paragraphs; mark anything needing a real number." Verify every figure — it may invent one.
4CFO business case"Proposal → 5 bullets for a cost-focused CFO, tied to ESG reporting, employer brand, or risk."
5Rehearse your boss"Act as my skeptical CFO — or General Counsel / CEO — give 5 objections, then pre-answer them."
6Stakeholder reply"Draft a reply to this rating-agency question in our cautious tone; flag what not to commit to in writing."
7Your AI position"Cautious one-paragraph internal AI line, EU AI Act-aligned, German Mittelstand voice." Align with Legal — not CC's alone to sign.
8Scope before you build"Partner says 'we want AI' → ask me 5 questions to bound it, and name the condition to say no."
5 · Get it approved — the pitch to steal

"The EU AI Act's staff AI-literacy duty has applied since Feb 2025 — we're likely non-compliant now — and the Act's penalty regime lands Aug 2026. Meanwhile staff already paste company data into private ChatGPT. I propose an 8-week pilot: 10 licenses of Copilot / Langdock for the CC team, measured on hours saved in CSRD reporting. ~€25/user/month. Owner: me."

Sources: artificialintelligenceact.eu (Art. 4) · zew.de (shadow-AI) · verify before quoting externally.

6 Point it outward — the social-impact lever

Program in 5 steps (Breakout 1 Track A builds it): scope the problem → match a partner → pilot-scope (days + liability boundary) → guardrail & Legal sign-off → measure as CSRD S3/S4.

Channel AI skills to an NGO

MI4People (KIISS) · N3XTCODER (AI for Impact) · CorrelAid (Data4Good)

Discounted tools for nonprofits

Microsoft for Nonprofits (Copilot −15%) · Salesforce Agents for Impact (pro-bono)

Free upskilling & EU AI Act help

KI-Campus · DRK-Lernplattform · vhs.cloud · ARIC Hamburg