Safety

Spark is an AI tutor for tech and AI fundamentals. It answers questions about how large language models work, how to write a good prompt, how the internet and HTTPS work, cybersecurity basics, and beginner coding.

Spark is not a friend, a therapist, or a companion. It is a computer program. It does not claim a relationship, it does not build a picture of a learner over time, and it is designed to be useful and then get out of the way. Before a learner’s first question, the app tells them plainly that Spark is an AI and not a person, and it repeats that during any unusually long sitting.

It does hold on to the conversation you are currently having. A follow-up question would be useless otherwise. Our server keeps the turns of an open conversation for up to 24 hours and then deletes them automatically, and closing the conversation in the app starts the next one from nothing. There is more detail on that below.

Spark can be confidently wrong. A line under the conversation says so the whole time it is open, and we would rather a learner check something than trust it.

If a message reads as self-harm, suicidal thinking, abuse, or acute distress, Spark does not answer it as a tutoring question and does not steer the conversation back to tech. The app immediately shows supportive text and real crisis resources:

• 988 Suicide and Crisis Lifeline. Call or text 988. Free, confidential, 24/7.
• Crisis Text Line. Text HOME to 741741. Free, confidential, 24/7.
• Chat at 988lifeline.org

Three things about how this is built. The phone numbers live in our code and are never written by the AI, so a wrong digit cannot reach a person who needs the right one. The check runs on our server, and it runs ahead of the usage limit and ahead of the check on our AI provider’s configuration, so a learner who has used every question in their plan, or who is asking at a moment when the tutor itself is not working, still gets the resources. And the response carries nothing but support and the resource: no disclaimer, no upsell, no advertisement for us.

Where that check does not reach. The check reads a message on its way to us, so it only exists on a path that sends one. If a learner has switched AI answers off, the app sends nothing at all, and there is nothing for the check to read. Rather than leave that gap unattended, the Ask Spark tab in that state has no question box in it at all, and it carries the 988 line directly on the screen. The message a learner would have typed is never captured, and the number is there anyway.

This is a signpost, not a crisis service, and it will not catch everything. It reads the words in a single message. It cannot read tone, it cannot read what is not written down, and no wording list catches every way a person says they are in trouble. If you are worried about someone right now, contact 988 or your local emergency services directly rather than relying on an app to notice.

Every tutoring answer Spark gives carries a Report control, right underneath the answer itself. Tapping it asks what went wrong (harmful or unsafe, inappropriate for my age, wrong information, off topic, or something else), lets the learner add anything they want to say, and sends it to us.

The one place that control does not appear is the crisis response. Nothing shares that screen with the phone number, by design, and there is nothing there for us to have got wrong: the words are ours, not the AI’s.

A report is not counted against a learner’s question allowance, and it is not blocked when that allowance runs out. Someone who has used every question in their plan can still report the answer that used the last one, because the learner most likely to have something worth reporting is exactly that person. There is a separate daily cap on how many reports one account can file, which exists only to stop the form being used to flood us. A person reads every report, and answers that should not have been given change what the tutor is allowed to say.

You can also email learnailesupport@gmail.com at any time.

In the app, open the You tab, find the AI section, and switch off AI answers. This is a real switch, not a hidden preference: with it off, the app makes no AI calls at all, so nothing a learner types is sent to our AI provider.

The Ask Spark tab stays where it is and says what happened, rather than disappearing or pretending to be live. There is no question box on it in that state, because a box that accepts a question nobody will answer collects a learner’s words for nothing. The 988 line sits on that screen, for the reason described above.

Lessons are unaffected, and so are the hints. Every graded step in the app ships with a hint we wrote ourselves, and Spark serves that written hint before it would ever call the AI. Those hints involve no AI and send nothing anywhere, so we leave them working rather than removing teaching in exchange for a privacy improvement you do not actually get.

When a learner asks Spark a question or requests an AI hint, that message is sent to our AI provider, Anthropic, to generate the reply.

Chat messages are kept for 24 hours, then deleted. Both sides of an Ask Spark conversation, what the learner wrote and what Spark replied, are stored on our server so the conversation holds together across turns. Every conversation carries a hard 24-hour expiry from the moment it starts. A job runs every hour and deletes what has expired, and expired turns are already filtered out before that, so nothing past its 24 hours is ever read back. We store the words because a tutor that forgets the previous sentence is not much of a tutor. We put a short clock on them because nothing about teaching someone how HTTPS works requires a permanent record of how they asked.

A crisis message is never stored at all. When the check described above fires, it fires before the conversation is written down and before anything is sent to our AI provider. That message is not saved on our server, it does not become part of the conversation history, and it never reaches Anthropic. Not for 24 hours. Not at all.

Separately, we keep a running count of requests and token totals on the account, for rate limiting and cost control. That record holds numbers, never words.

Reports are the one thing we keep for longer, because a report is a request for a person to look at something. When a learner reports an answer, we store the answer, anything they typed into the report, and the question that produced it if they leave that box ticked. There is no way to review a report without the thing being reported. If a reported message reads as a crisis disclosure, we drop the learner’s words and keep only the answer: a report must never turn into a permanent record of a young person’s worst moment. Reports are kept until a person has reviewed them, and for a year after that, so a pattern across several reports is still visible months later.

Deleting an account takes the progress, any conversation still inside its 24 hours, the reports, and the usage counts. It is in the app, it happens immediately, and it cannot be undone. Two things survive it, and neither can be traced back: the safety-event labels described below, which are kept with the account reference stripped off, and the billing record we have to keep for a purchase that was made.

We also count safety events, so we can tell whether the controls on this page are actually working. When a filter declines a request or blocks a reply, we record short labels for what kind of event it was against the account, never the text of any message. Moments where we show crisis resources are counted only as a daily total, with no account attached and no text: we can see that the signpost was needed on a given day, and we cannot see who needed it.

The full detail is in our Privacy Policy, and the list of companies that process data on our behalf is at Subprocessors.

AILE is intended for ages 13 and up. The app asks for confirmation on first launch and does not knowingly collect data from children under 13. If you believe a child under 13 has created an account, email learnailesupport@gmail.com and we will delete it.

We treat every learner as though they may be a minor. The AI disclosure and the take-a-break reminder are shown to everyone rather than only to users we have confirmed are young, because the cost of getting that wrong in the other direction is a teenager treated as an adult.

If something about the tutor worries you, we would rather hear it than not. Email learnailesupport@gmail.com. AILE LLC, 9251 West 22nd Street, Minneapolis, MN 55426.