AI & Automated Processing Statement

Core principle: Mail-Organiser's AI classifies email metadata. It never reads email content. No automated action is ever taken without your explicit approval. You are always in control.

1. Overview

Mail-Organiser uses artificial intelligence to analyse and classify your email metadata. This statement explains how our AI systems work, what data they process, how decisions are made, and how you can understand, challenge, or override any automated classification.

This statement is provided in accordance with UK GDPR Articles 13, 14, and 22, which require transparency about automated processing and meaningful human oversight of any automated decisions that significantly affect individuals.

2. What Our AI Does

2.1 Email Classification

When you run a scan, Mail-Organiser's classification AI analyses the following metadata for each email:

The sender's email address and display name
The email subject line
The date and time the email was received
The current folder location

Based on this metadata, the AI assigns each email to one of 14 predefined categories (Banking, Health, Shopping, Newsletters, etc.). The AI does not read the body content of any email.

2.2 Inbox Score

The Inbox Score (0–100) is a computed metric based on the ratio of categorised to uncategorised emails, the distribution of categories, and trends over time. It is a summary metric, not a judgment about the quality of your emails.

2.3 Sentinel Detection

The Sentinel system analyses email metadata for patterns consistent with prompt injection attacks — attempts by malicious senders to embed AI instructions in email subject lines. Emails flagged by Sentinel are classified as suspicious and never processed further.

2.4 MO Chat

The MO Chat feature allows you to ask questions about your inbox management. Responses are generated by Claude (Anthropic) based on your question and general knowledge of email organisation. Chat responses do not access your actual email content or personal data beyond what you share in the conversation.

3. The No-Auto-Action Principle

This is one of Mail-Organiser's core design commitments:

No email is ever moved, deleted, or modified without your explicit approval.

Every scan produces a results screen. You must click "Approve All" or individually select actions before any email is moved. You can review each result before acting.

Protected categories are never moveable by any automated process.

Banking, health, HMRC, and flagged important emails are shown in results but are excluded from all move operations, even if you click "Approve All".

You can undo any approved batch.

After approving a batch, the "Undo Last Batch" button returns all moved emails to your inbox. This is available until you run a new scan.

4. UK GDPR Article 22 Compliance

Article 22 of UK GDPR provides protections against solely automated decisions that produce legal effects or similarly significant effects on individuals.

Mail-Organiser's email classification does not produce legal effects. The classification of a newsletter as "marketing" does not constitute a legal decision about you. However, we go beyond the minimum legal requirements by:

Providing a transparent explanation of the classification for every email scanned
Never taking automated action — all consequences require your explicit approval
Providing the ability to override any classification by simply not approving the move
Providing an "Undo" function after any approved batch
Logging all actions in the activity feed for your review

If you believe an email has been misclassified, you simply do not approve the move for that email. Protected categories provide an additional safeguard for the most sensitive email types.

5. The AI Models We Use

Mail-Organiser's classification uses models provided by Anthropic (Claude). Key facts about our AI usage:

Input to AI: Email metadata only (sender address, subject line, timestamp). Email body content is never sent to the AI.
No training on your data: Your email metadata is not used to retrain AI models. Classification is performed by the model at inference time only.
Anthropic's data use: Anthropic processes the metadata we send solely to provide classification responses. We have contractual protections against Anthropic using this data for model training. See Anthropic's Privacy Policy for details.
Classification accuracy: AI classification is not perfect. The system is designed to be conservative — when uncertain, the AI is more likely to leave an email unclassified than to misclassify it into a protected category.

6. Your Rights Regarding Automated Processing

Right to explanation: You can see the category assigned to each email in the scan results screen.
Right to challenge: You can override any classification by not approving a move, or by undoing a batch after approval.
Right to opt out: You can disable the monitoring feature in Settings. You can also choose not to run scans.
Right to deletion: You can delete your account and all associated classification data at any time.

7. Bias and Fairness

We recognise that AI classification systems can exhibit bias. We monitor our classification system for patterns that might indicate systematic misclassification for particular types of senders or email patterns. Our UK-specific training recognises UK domain patterns, institutions (NHS, HMRC, Royal Mail), and financial institutions to minimise misclassification for UK users.

If you notice a pattern of systematic misclassification, please report it to ai-feedback@mail-organiser.com. We investigate all credible reports of systematic AI error.

8. Changes to AI Systems

We will notify users of material changes to our AI systems — including changes to classification criteria, new categories, or changes to the AI provider — via email or in-app notification at least 14 days before changes take effect.

9. Contact

For questions about our AI systems: ai-feedback@mail-organiser.com
For data protection matters: privacy@mail-organiser.com