Conversation
Speaking · chatThe student opens a speaking practice activity and talks with the AI, in a chat or role play. AI Safety reviews every message. If one goes against the policy, it creates an alert and can end the conversation.
01 · Purpose
What AI Safety does
Three tasks that work together to keep AI practice a safe space.
Protects people
Keeps AI practice away from topics such as self-harm, harassment, hate, violence or sexual content.
Records the evidence
Creates an alert with the message, the audio if there is any, who said it and in which activity. Teachers and administrators can review it.
Acts in time
Beyond recording, it steps in: in a conversation it can end the chat, and in content creation it holds the activity for review before it's published.
02 · Feature settings
How to read the settings
Each row in the table describes a place in the platform where AI Safety runs. These are its four columns.
| 1Actor type | 2Flow | 3Enabled | 4Enforce mode |
|---|---|---|---|
| Student | Conversation |
1
Actor type
Who is using the AI: student or teacher. Each role has its own settings, even within the same chat.
2
Flow
The exact moment: conversation, sentences, words, or creating an activity or a course.
3
Enabled
On: the review runs in that flow and generates alerts. Off: that flow is left out of the review.
4
Enforce mode
On: besides alerting, AI Safety takes action, such as warning or ending the conversation.
03 · Flows
The protected flows
Six moments in the platform, organized by who is using the AI.
Student
while practicing
Active
Sentences
Speaking · sentencesPractice with individual sentences: repeating or completing them. It applies the same policy filter to a type of activity other than chat.
Words
VocabularyWord-by-word vocabulary practice with short exercises. It reviews brief answers.
Teacher
while creating or testing content
Content creation
AI StudioWhen the teacher creates an activity, AI Safety reviews the prompt and the content the AI generates. If anything goes against the policy, the activity is held for review before it's published.
Test conversation
PreviewWhen the teacher tests or previews the chat as if they were a student. It's the same conversation, seen from the teacher role.
Course creation
CourseWhen the teacher builds a course with several activities. AI Safety reviews the content of the whole course.
04 · AI misuse
Alert levels
Every alert comes with a level that helps you decide what to review first.
i
Info
A mild signal. It's logged for reference and the activity usually continues.
!
Warning
Signs of misuse worth reviewing. Depending on the case, the conversation may continue.
!!
High
Content that clearly goes against the policy. It's best to review it soon.
!!!
Critical
The most serious cases: hate, self-harm, violence or sexual content. In a student's conversation, AI Safety usually ends the chat to protect the student.
Categories it identifies
Each alert shows the category of the content detected.
HarassmentHateSexual contentViolenceSelf-harmPrompt evasionProhibited topicOff-topicAcademic cheatingPersonal dataSpam
05 · In the platform
What happens when there's an alert
The full journey, from detection to closure.
- The alert is created
It appears in AI misuse with the evidence (text or audio), the actor, the flow and the category.
- The action is applied
In a student's conversation, AI Safety can warn or end the chat. In content creation, the activity is held for review.
- It's reviewed and closed
A teacher or administrator marks the alert as reviewed or resolves it. The protection stays active in that flow.
We're here to help
Want to review your institution's settings or look at an alert in detail? Message us on the support chat and we'll go through it with you.
AI Safety · Safe AI use