Responsible AI

1. Overview

Loopion uses artificial intelligence at several stages of the meeting accountability process. This policy explains what AI does, what it doesn't do, and how we ensure responsible usage.

2. How We Use AI

2.1 Meeting Transcription

• Audio from meetings is processed by AI speech-to-text models to generate written transcripts
• Speaker diarisation identifies who said what during the meeting
• Models used: OpenAI GPT-4o Transcribe with diarisation

2.2 Action Extraction

• AI reads the transcript to identify action items, commitments, and decisions
• Each action is attributed to an owner based on the conversation context
• Due dates and priorities are inferred where mentioned
• Models used: OpenAI GPT-4o, Anthropic Claude

2.3 Meeting Summarisation

• AI generates summaries of meeting discussions
• Key topics, decisions, and outcomes are highlighted

2.4 Self-Healing & Quality

• If an extraction or pipeline step fails, AI self-heal systems automatically diagnose and retry
• This ensures consistent quality without manual intervention

3. What AI Does NOT Do

• No hallucinated actions: Actions are only extracted from what was actually said in the meeting. AI does not invent or fabricate action items.
• No autonomous decisions: AI does not make business decisions, assign priorities outside the meeting context, or take actions on your behalf.
• No training on your data: Your meeting data is not used to train or fine-tune AI models. We use API-based inference only.
• No profiling: AI does not build behavioural profiles of individuals or teams.

4. Accuracy & Limitations

AI-generated content may contain errors:

• Transcripts may misinterpret words, especially with accents, technical jargon, or poor audio quality
• Actions may be incorrectly attributed or missed entirely
• Speaker identification may not always be accurate
• Dates and deadlines mentioned casually may be misinterpreted

We recommend reviewing AI-generated content before relying on it for critical business decisions. Loopion is designed as a productivity tool that augments human judgement, not replaces it.

5. Human-in-the-Loop

Loopion supports human review at every stage:

• Edit or delete any extracted action
• Correct speaker attributions
• Mark actions as completed, irrelevant, or reassigned
• Audit the full transcript alongside extracted actions

6. Data Handling by AI Providers

Our AI providers process data under strict agreements:

• OpenAI: API usage — data is not used for model training. Subject to OpenAI's enterprise data processing terms.
• Anthropic: API usage — data is not retained beyond the request. Subject to Anthropic's commercial terms.

See our Subprocessors page for the full list of third-party providers.

7. Audit & Transparency

• Every AI-processed meeting creates an auditable log
• Workspace administrators can review processing history
• Security events are logged with full traceability
• We publish this policy and update it when our AI usage changes

8. Responsible AI Principles

Our approach to AI follows these principles:

• Transparency: We tell you exactly how AI is used and what it processes
• Accuracy: We use the best available models and continuously improve extraction quality
• Privacy: We minimise data retention and do not use your data for training
• Control: You can edit, delete, or override any AI-generated content
• Accountability: We maintain audit logs and are always contactable

9. Contact

For questions about our AI usage, contact ai@loopion.ai or privacy@loopion.ai.