AI Usage Disclosure

1. What AI Does on This Platform

Noryn uses AI models to power the following capabilities:

• Message generation: Drafting replies to inbound messages from your customers' contacts, in the voice and style you configure
• Reply classification: Categorising inbound messages by intent (inquiry, complaint, booking request, etc.)
• Lead follow-up: Generating outbound follow-up messages based on your configured templates and context
• Scheduling assistance: Suggesting and confirming appointment slots based on calendar availability
• FAQ handling: Answering common questions using content you provide
• Conversation routing: Deciding when to escalate a conversation to a human

2. What AI Does NOT Do on This Platform

• Make decisions with legal or regulatory effect (hiring, lending, insurance, legal proceedings)
• Provide regulated professional advice (medical, legal, financial) without a domain-specific addendum
• Replace human judgment — all output is advisory and requires human review before reliance
• Store any information independently of the platform — the AI model receives only what the platform sends in each request
• Learn from your data for the benefit of other customers (see section 5)

3. AI Models Used

We use Anthropic's commercial API under their Commercial Terms of Service, which includes a Data Processing Agreement. Anthropic does not train on data sent via the commercial API.

4. What Data Is Sent to the AI

Each AI request includes only what is necessary for the task:

• The system prompt you configured for the Virtual Employee
• The inbound message content
• Relevant conversation history (up to the configured context window)
• FAQ or knowledge base content you provided
• Voice samples and tone examples you submitted during setup
• The current date and time (for scheduling)

5. What Is NOT Sent to the AI

• Your billing or payment information
• Account passwords or authentication tokens
• Integration OAuth credentials
• Data belonging to other customers
• Your own personal data beyond what you include in your system prompt

6. AI Training and Data Retention by Anthropic

Under Anthropic's Commercial Terms of Service, Anthropic does not use data submitted via the API to train or improve their models without explicit permission. Your conversation data is processed by Anthropic's infrastructure to generate a response and is not retained for training purposes.

Anthropic's privacy information: privacy.anthropic.com

7. Your Obligations Regarding AI Transparency

7.1 Disclosure to your end-users

UK ASA guidance and emerging AI transparency requirements expect that individuals interacting with AI-powered communications are informed they may be doing so. We require:

• Standard and Pro tiers: All outbound AI-generated messages include a footer: "Sent on behalf of [Business Name] using AI assistance." This cannot be removed.
• Premium and Ultra tiers: The footer can be disabled in employee settings, but doing so transfers full legal accountability for AI disclosure to you. An in-platform warning explains this before you make the change.

7.2 Human review obligation

You are responsible for reviewing AI-generated communications before relying on or distributing them, especially where the content could have legal, financial, health, or significant personal consequences for the recipient.

7.3 Regulated industries

If you operate in healthcare, legal services, financial services, or other regulated industries, you must ensure your use of AI-generated communications complies with all sector-specific regulations (e.g., FCA rules, SRA guidance, CQC requirements). We offer domain-specific addenda for Premium and Ultra customers — contact us to arrange this before deploying Virtual Employees in these contexts.

7.4 Human takeover

When a human takes over a conversation from a Virtual Employee, outbound messages switch to a "Sent by [Your Name]" indicator. Do not use AI-generated content in this mode without re-enabling the Virtual Employee — doing so could mislead your contact about who they are communicating with.

8. Limitations of AI Output

AI language models, including Claude, can:

• Produce factually incorrect information presented confidently ("hallucination")
• Reflect biases present in their training data
• Generate inappropriate output in edge cases
• Misunderstand ambiguous instructions or context
• Fail to update their knowledge of recent events

This is why human oversight is not optional — it is a requirement of responsible AI deployment.

9. Feedback and Improvement

If a Virtual Employee produces output that is harmful, factually wrong, or inappropriate, please report it via the conversation feedback button in your dashboard or email privacy@norynsystem.com. We use this feedback to improve system prompts and configurations, not to retrain the underlying AI model.