Safeguarding

ChemResults is used by students aged 13 to 18. Keeping them safe is built into how ChemResults works, not added afterwards. This page explains the safeguards around ChemChat, our AI tutor. How we handle personal information is explained in our Privacy Policy.

Teacher-approved feedback comes first

When a student answers a question, the feedback they see first has been written, or reviewed and approved, by teachers. It is prepared and checked in advance, not generated by AI at the moment the student answers, so every student who chooses the same answer sees the same feedback. Teachers can always see exactly what their students are learning from. ChemChat only answers follow-up questions, and it answers them in the context of that question and its teacher-approved feedback.

ChemChat only talks about chemistry

Before ChemChat replies, a separate check decides whether the message is about chemistry. If it isn't, ChemChat politely declines and invites the student to ask a chemistry question instead.

Every message is checked, both ways

We have built and tested several layers of safety checks, called guardrails, around ChemChat.

What a student types is checked before ChemChat answers. A message is blocked if it contains:

• harmful or inappropriate content: violence, sexual content, hate, insults, or encouragement of criminal or dangerous activity
• bad language
• an attempt to trick ChemChat into ignoring its instructions (a "prompt attack")
• personal details, such as a phone number, an email or home address, a password, or bank or card details

Every reply from ChemChat is checked in the same way before the student sees it. A reply that fails the check is never shown.

When something is blocked, the student sees a message saying it was blocked under our Safeguarding Policy, with a link to this page.

We test these checks against real chemistry, so that genuine questions, such as those about explosive reactions, toxic gases or famous chemists like Le Chatelier and Mendeleev, are still answered.

Replies are displayed safely

ChemChat's replies are not web pages written by the AI. They are structured text that ChemResults itself displays, so a reply can't contain hidden code. A reply can only include a link if it comes from the question's own reference sources.

Conversations are private and not stored

• A ChemChat conversation exists only in the student's open browser tab. It is gone when they leave the question or close the tab. We don't store it, and we have turned off the option for Amazon Web Services (AWS), which runs the AI model, to store it.
• Messages are processed in the Asia-Pacific region and are not used to train AI models.
• The student's email address, username and account ID are never sent to the AI model.
• ChemChat has a daily message limit.

We keep improving

We monitor how often each safety check is triggered, recording only which kind of check it was, never what was typed. That lets us spot when a check blocks genuine chemistry, or misses something it should catch, and improve it.