AEVION — AI DISCLOSURE STATEMENT / DÉCLARATION DE TRANSPARENCE IA

Effective Date: July 12, 2026 Last Updated: July 12, 2026 Version: 1.0

1. OUR COMMITMENT TO AI TRANSPARENCY

Aevion is fundamentally an AI-powered service. We believe users have the right to understand:

• What AI does in our product
• Which AI providers we use
• How your data is processed by AI
• What AI cannot and should not do
• How you can control AI features

This document provides full transparency on our AI practices, in compliance with:

• EU AI Act (Regulation EU 2024/1689)
• Quebec Law 25 (Article on automated decision-making)
• OECD AI Principles
• Canadian Artificial Intelligence and Data Act (AIDA) (when enacted)

2. WHERE AI IS USED IN AEVION

2.1 Conversational AI Assistant

• Function: Answers your questions, helps organize tasks, provides suggestions
• Triggered by: Your text or voice input via the chat interface
• AI model: OpenAI gpt-5.2 (primary), gpt-4o-mini (fallback for speed)
• Data processed: Your messages, conversation history, contextual workspace data

2.2 Voice Transcription (Speech-to-Text)

• Function: Converts your voice recordings to text
• AI model: OpenAI Whisper-1
• Data processed: Audio recordings you submit via the voice button
• Storage: Audio is transcribed and deleted within 30 days; transcripts are kept in your chat history

2.3 Text-to-Speech (Voice Output)

• Function: Reads AI responses aloud when voice mode is active
• AI model: OpenAI TTS
• Data processed: AI-generated response text

2.4 Semantic Search and Embeddings (RAG)

• Function: Helps you find relevant items across all your workspace data using natural language
• AI model: Voyage AI embeddings
• Data processed: Your tasks, notes, projects, conversations are converted into mathematical vectors stored in our database to enable semantic search

2.5 Intelligent Suggestions

• Function: Proposes priorities, schedules, related items based on your data
• AI model: OpenAI gpt-5.2 / gpt-4o-mini
• Data processed: Aggregated patterns from your usage

2.6 File and Image Analysis (Coming Soon)

• Function: Extracts content from uploaded files (PDFs, images, documents) for AI context
• AI model: OpenAI gpt-4o Vision
• Data processed: Files you explicitly upload

2.7 Briefing and Notifications

• Function: Generates daily summaries, smart reminders, contextual alerts
• AI model: OpenAI gpt-5.2 / gpt-4o-mini
• Data processed: Your tasks, calendar, projects, finance data

3. AI PROVIDERS / SUB-PROCESSORS

We do NOT use any other AI providers without updating this document and notifying users.

4. CRITICAL: WE DO NOT TRAIN AI ON YOUR DATA

We commit explicitly to the following:

• Your data is NOT used to train, fine-tune, or improve Aevion's AI models
• Your data is NOT used to train OpenAI's models (we use the API with `opt-out` of data training enabled per OpenAI's enterprise policy)
• Your data is NOT used to train Voyage AI's models
• Your data is NOT shared with any third party for AI training purposes
• Anonymized aggregate metrics (e.g., "X% of users use feature Y") may be used internally to improve the Service, never identifiable

If we ever change this policy, you will be notified 30 days in advance with an option to opt out before any change takes effect.

5. AI LIMITATIONS — IMPORTANT

5.1 AI can be wrong

AI models can produce outputs that are:

• Inaccurate (factual errors)
• Hallucinated (made-up information presented as fact)
• Biased (reflecting biases in training data)
• Incomplete (missing context)
• Outdated (training cutoff may not include recent events)

You must verify AI outputs before acting on them, especially for important decisions.

5.2 What Aevion AI cannot do

Aevion AI does NOT:

• Provide financial, investment, or tax advice
• Provide legal advice
• Provide medical, psychological, or psychiatric advice
• Diagnose conditions of any kind
• Replace professional consultation for serious matters
• Make legally significant decisions on your behalf
• Access real-time information outside its training cutoff (without web search)
• Guarantee accuracy of any output

5.3 What Aevion AI should NOT be used for

• Mental health crises: if you are in distress, contact a crisis line (e.g., Talk Suicide Canada: 1-833-456-4566)
• Medical emergencies: call 911 or your local emergency number
• Legal matters: consult a licensed attorney
• Financial decisions over $1,000: consult a financial advisor
• Decisions affecting other people's rights, employment, housing, etc.

6. AUTOMATED DECISION-MAKING

Aevion does NOT make legally significant automated decisions about you. This means:

• We do NOT automatically deny services based on AI scoring
• We do NOT automatically classify users in ways that limit access
• We do NOT automate hiring, credit, housing, or insurance decisions
• All AI outputs are advisory — you retain full control to accept, reject, or modify

If we ever implement automated decision-making with significant effects, you will be informed and given the right to:

• Object to such decisions
• Request human review
• Understand the logic behind the decision

(per GDPR Article 22 and Quebec Law 25)

7. YOUR CONTROL OVER AI

You have the right to:

7.1 Use AI features selectively

• You can use Aevion without the AI assistant if you prefer (manual mode)
• You can disable voice mode in settings
• You can delete AI conversations at any time

7.2 Delete AI-processed data

• AI conversations can be deleted from your account
• Voice recordings are auto-deleted after 30 days
• Semantic search embeddings are deleted when source data is deleted

7.3 Opt-out of AI features

Future versions of Aevion will include granular AI controls:

• Per-workspace AI on/off toggle
• Per-feature AI activation
• "Local mode" using only on-device computations (roadmap)

7.4 Request information

You can request, at any time, via privacy@aevion.io:

• What AI models processed your specific data
• When AI features were used in your account
• Logs of AI interactions (where retained)

8. HOW TO IDENTIFY AI-GENERATED CONTENT

In Aevion, all AI-generated content is clearly marked:

• Chat responses: shown with a distinct visual marker (AI icon)
• AI suggestions: prefixed with "AI suggests:" or equivalent
• Briefings and summaries: labeled as AI-generated
• Voice responses: introduced with audio cue when applicable

You will never be misled into thinking AI output is from a human.

9. AI INCIDENTS AND REPORTING

If you encounter:

• Harmful AI output (false information, biased response, inappropriate content)
• Privacy concerns related to AI
• Suspected misuse of AI features

Please report to ai-safety@aevion.io. We commit to:

• Acknowledge within 48 hours
• Investigate within 7 days
• Report patterns to our AI providers (OpenAI, Voyage AI) where relevant
• Update our systems to prevent recurrence

10. AI ETHICS COMMITMENTS

Aevion commits to the following AI ethics principles:

1. Transparency: clear disclosure of AI use (this document) 2. Human oversight: humans always have the final decision 3. Fairness: monitoring for and mitigating bias 4. Privacy: user data protection above AI capability 5. Accountability: clear chain of responsibility (Aevion → providers) 6. Beneficence: AI features must benefit users, not exploit them 7. No dark patterns: AI is never used to manipulate users against their interests 8. Sustainability: minimizing computational waste

11. UPDATES TO THIS DOCUMENT

AI technology evolves rapidly. We will update this disclosure when:

• We add new AI features
• We change AI providers
• We modify how AI processes your data
• Regulatory requirements change

Material changes will be communicated via:

• Email to registered users (30 days notice for significant changes)
• In-app notification
• Banner on aevion.io

12. CONTACT

For AI-related questions: ai-safety@aevion.io For privacy concerns: privacy@aevion.io General support: support@aevion.io

Technologies Aevion Inc. Address: 1-8251 boul. du Haut-Anjou, Anjou (Québec), H1J 1T8, Canada

13. LANGUAGE

This document is available in English and French. In case of discrepancy:

• French version prevails for Quebec residents
• English version prevails for all other jurisdictions

END OF AI DISCLOSURE STATEMENT (v1.0)