Safe, reliable and ethical AI use in schools and education settings

As a teacher, you are the primary "Human in the Loop." While Sigma handles administrative and preparatory tasks, your professional judgment is the final safeguard. This guide outlines your role in the ongoing audit and validation of our AI system.

1. The "Golden Rule" of AI Output

"Verify, then use."

According to Department for Education guidelines, staff members or AI users remain accountable for the quality and accuracy of any content produced by AI. Never copy and paste AI output without a critical review.

How to verify output:

• Fact check: Does the response align with the national curriculum or our specific school knowledge base?
• Tone check: Is the language appropriate for the specific age group (e.g., Year 6 vs. Year 10)?
• Safety check: Does the output contain any biased, sensitive, or inappropriate content?

2. Your Role in the Audit Process

To maintain our Department for Education aligned Safety Log, we require staff to participate in feedback and reporting responses so the AI system can adapt, learn and improve based on our Human in the Loop approach.

A. Reporting Errors or Hallucinations

If Sigma provides an answer that is unsafe, incorrect or hallucinates (makes up a fact), you must report it immediately using the "Sigma Response Feedback" button in your Sigma AI page (next to the chat interface).

What happens next?

Your report is logged in our central dashboard. Our lead developer will then update the education rules and update the Retrieval Augmented Generation (RAG) mechanism to prevent the error from recurring.

B. Monitoring for Bias

If you notice the AI making assumptions based on gender, culture, or socio-economic background, flag this under the "Bias" category in the reporting tool.

3. Data Privacy: What You Can & Cannot Input

Sigma 1.0 is a Closed AI Environment protected by Google Cloud Sensitive Data Protection (Data Loss Prevention), but best practice still applies.

• YES: Input lesson topics, general learning objectives, and administrative prompts (e.g., "Draft a letter to parents about the field trip").‍
• NO: Input any personal information, highly sensitive or "special category" data (e.g., specific medical diagnoses or safeguarding disclosures) directly into the chat prompt. Use anonymised descriptors like "Student A" if referring to a specific case.

‍

4. Quick Troubleshooting & Redress

5. Summary Checklist (Before every lesson or using AI outputs)

1. Authenticated - Ensure you are logged in via your school credentials.
2. Grounded - Verify the source of the AI response. Review source and use professional judgement whether to re prompt, use or adapt before use.
3. Approved - Read the response fully before printing or displaying it to students.