AI Ethics Policy

1. Purpose

To explain where, with which providers and within which limits we use artificial intelligence at hangel, and how we ensure human oversight, transparency and the protection of personal data.

2. Scope

All AI-supported features on the hangel platform and the AI tools the Association uses in its own work.

3. Principles

1. People decide. AI produces drafts and suggestions; decisions that affect people (report decisions, organisation approval, certificates, payouts) are taken by people.
2. Transparency. Our rule is that content generated or translated by AI is indicated as such; today, translations have a "show original" option and application drafts carry an AI use declaration.
3. Responsibility for accuracy. AI output can be wrong. Users read and correct it before using it; the system never declares "read and verified" on a user's behalf.
4. Data minimisation. We aim to send AI only the data the task needs. Blood group and health consent records are not inputs to AI features; however, health information may come up in free conversation, such as call-centre calls, and may be transcribed.
5. Fairness. AI is not used to rank or filter people by religion, ethnic origin, disability or similar characteristics.
6. Reference framework. We follow the transparency and human-oversight principles of the EU AI Act (2024/1689) and the UNESCO Recommendation on the Ethics of AI; the EU Act does not apply directly in Türkiye.

4. Roles and responsibilities

• Board of Directors: approves new areas of AI use.
• Super-admins: manage provider keys and model choice.
• Product team: designs data flows, labelling and human oversight for new features.
• Users and organisations: are responsible for checking AI output.

5. How it works at hangel today

• Providers: The main provider is Google Gemini; Anthropic Claude can also be used for project writing. Provider keys are kept in records that only the server can read and are never shown in full on any screen.
• Areas of use:
  • Write Your Project: drafting grant and funding applications, scoring and improvement suggestions,
  • translating organisation content into the visitor's language (with a "show original" option),
  • creating an event draft from a single sentence (with a daily quota per organisation),
  • extracting profile information from a CV uploaded by a volunteer,
  • transcription and summaries of call-centre conversations,
  • advertising and campaign planning assistants.
• Contribution trail and declaration: Write Your Project calculates, from the version history, the difference between the text written by AI and the user's edits for each section. The AI use declaration to be attached to an application lists the models used and the data types actually sent. The "read and verified" statements are added only if the user confirms them personally.
• Personal data: In application drafts, managers' names and contact details from the organisation file may be sent to the AI; this is stated openly in the use declaration.
• Abuse and cost limits: User inputs are stripped of control characters and length-limited; a daily quota applies per person and task type.
• Moderation: Content reports are not decided by AI; a person reviews them.

6. Not yet in place and our goals

• No written risk and impact assessment has been carried out for AI features; our goal is a short assessment template for new features.
• There is no regular check for discrimination in recommendation and ranking results.
• The legal review of providers' contractual terms on whether submitted data is used for model training has not been completed.
• The existing "AI and Algorithmic Transparency Statement" does not cover the generative AI uses listed above; it needs updating.
• Automatic labelling of AI-generated images is not available on every surface.

7. How to raise a concern

Errors, bias or complaints about AI output: destek@hangel.org. Objections to processing of personal data with AI: kvkk@hangel.org.

8. Review

At least twice a year and before any new AI feature or provider is added.

9. Related documents

Existing "AI and Algorithmic Transparency Statement" · Personal Data Protection Policy · Language and Translation Policy · Supplier and Third-Party Policy