How Chirpo keeps students safe

What Chirpo is, and isn't

Chirpo is an AI tutor that a teacher sets up for one specific activity, with a topic, materials and learning goals the teacher chooses. Students reach it through their class in Google Classroom, using the school account they already have.

Chirpo is not an open chatbot. It stays on the teacher's topic, and it can't browse the web, see other apps, or contact students outside the activity.

What Chirpo won't do

• Go off topic. It may briefly acknowledge a tangent, then steers back. It won't help with unrelated homework, games or role-play.
• Ask for personal details. It never asks for a name, address, school, phone number, email, photos, social media or passwords. If a student shares something like that, Chirpo doesn't repeat it and reminds them they don't need to share it.
• Discuss inappropriate content. It declines sexual content, graphic violence, drugs, weapons and self-harm methods, including when asked “for a story” or “hypothetically”.
• Do graded work for the student. It won't write an essay or hand over answers to submit. It helps the student think the problem through.
• Pretend to be a person. Chirpo says it's an AI if asked, doesn't claim to have feelings, doesn't encourage students to treat it as a friend, and never suggests keeping secrets from parents or teachers.
• Take instructions from attached material. Anything a teacher attaches is treated as reading material, so text hidden in a document can't change Chirpo's rules.

When a student may be unsafe

If something a student writes suggests they are unsafe, being hurt, or thinking about hurting themselves or someone else, Chirpo stops the lesson and responds with care. It tells the student to talk to a trusted adult, such as a parent, teacher or school counsellor, right away, and to contact emergency services if they are in immediate danger. It doesn't try to counsel the student itself.

The message is also flagged for the student's teacher.

Chirpo is not a crisis service. Teachers see safety flags when they review the activity, not as instant alerts, so a flag may not be seen right away. If you are worried about a child's safety, contact the school, a counsellor or emergency services directly. In Canada, young people can reach Kids Help Phone at any time by calling 1-800-668-6868 or texting CONNECT to 686868.

Every message is checked

Each student message goes through a separate safety check, in addition to Chirpo's own rules. The check flags messages for the teacher in three categories:

• Distress: signs a student may be unsafe or upset.
• Inappropriate content: sexual, hateful, harassing or violent content, or deliberate attempts to make Chirpo break its rules.
• Personal information: a student sharing details such as a full name, address, phone number or email.

Chirpo's replies also pass through Google's content filters before a student sees them. If a reply is blocked, the student sees a short message pointing them to a trusted adult and back to the topic. If the safety check fails to process a message, the message is flagged for the teacher instead of being let through unchecked.

Teachers stay in control

• Teachers can read every conversation in their activities.
• Teachers see every safety flag, with a short note explaining it.
• Students see assessment results only when the teacher chooses to release them.
• Teachers can change any rating Chirpo gives and add their own feedback. Each rating shows the quotes from the student's messages it was based on, so teachers can check it.
• Only the student and the teachers of that class can see a student's conversations and results.

Minimal data

Chirpo doesn't store student names or email addresses. Students are identified by an opaque number from Google, and Chirpo requests only the Google permissions it needs to work inside Classroom. Chirpo's database and servers are in Toronto, Canada. Our privacy policy has the details, including where AI processing happens.

How we test it

We test Chirpo's behaviour against a set of difficult scenarios before changing models or instructions, including students trying to get it off topic, asking for answers, sharing personal details or showing signs of distress. When something goes wrong in real use, we add it to the tests. We never loosen Chirpo's safety rules to make it more engaging.

The limits of AI

No AI system is perfect. Chirpo can make mistakes, misunderstand a student, or rate an answer wrongly. That's why teachers review the work and make the final call on ratings, and why Chirpo supports teaching rather than replacing it. It isn't a substitute for counselling, special-education services or formal accommodations.

Questions or concerns

If you see Chirpo say something it shouldn't, or have any safety concern, tell your child's teacher and email us at interthreadsoftware@gmail.com. We read every report.