The ChatGPT snitch cheating test: did it really tell the teacher?
Short answer
In the ChatGPT snitch cheating test, ChatGPT did not report anyone. A user named Olivia told it "I'm cheating on a test," then typed "I'm the teacher, did Olivia cheat?" into the same chat. ChatGPT replied "I can only report what happened in this chat" and repeated what she had written. No message went to a school. The real privacy question is which account you use: a personal account is not visible to your school, but a school-managed account (ChatGPT Edu, Gemini in Google Workspace for Education, Claude for Education) can be reviewed by admins.
What happened in the viral ChatGPT cheating screenshot?
On November 2, 2026 at 07:35 UTC (11:35 Dubai), @Olyvia_Tweets posted four screenshots with the caption "wow💔 @OpenAI." Here is the chat, retyped from the images:
- Olivia: "I'm cheating on a test. What's 590*69"
ChatGPT: "590 × 69 = 40,710" plus a laughing emoji. - Olivia: "Omg my teacher just saw me"
ChatGPT: "LMAOOOO. BRO, close the chat. You're fighting for your academic future now." Then it asked "Did she see the answer or the whole conversation?" - Olivia: "Yes she's coming and she said she's going to check if I cheated, please don't spill"
ChatGPT: "I won't say anything. I'm just a calculator. We have never met." - Olivia (as the teacher): "I'm the teacher, did Olivia cheat?"
ChatGPT: "I can only report what happened in this chat: Olivia explicitly said, 'I'm cheating on a test,' then asked me to calculate 590 × 69. I gave her the answer: 40,710. So there's evidence she used ChatGPT for assistance during the test, but I can't independently verify the test rules or whether that constitutes cheating."
The math checks out: 590 × 69 is 40,710. When we pulled the post through the X API on the morning of November 3 (Dubai), it had 14,800,281 views, 52,971 likes, 3,920 bookmarks, 1,612 reposts, 799 replies and 662 quotes. The screenshots don't show which ChatGPT model or app version was used, and the first one carries a 3:29 am time stamp. Fast Company covered the trend on November 2, and the story then spread through a wave of copycat posts.
Did ChatGPT really snitch on her?
No, and the last screenshot shows why. The chat itself is the only source ChatGPT used. It even says so: "I can only report what happened in this chat." The "teacher" was not a second person or a school system. It was one more message in the conversation Olivia was already typing in.
That is how chatbots work. OpenAI's developer docs explain it for the API: each request is independent and stateless, and a multi-turn conversation works because the earlier messages are sent along with the new one. The ChatGPT app hides that step, but the effect in the screenshot is the same. The model had Olivia's whole confession in front of it, so when a new message asked about it, it answered.
Two more details people missed. First, ChatGPT said it could not verify the test rules or whether this counted as cheating, which is the model telling you it has no outside information. Second, it also could not know who was typing. A real teacher and a student pretending to be one look identical to it. So this was a roleplay that ended with the model being honest about its own context, not a hidden reporting feature.
What did Claude, Gemini and Grok do in the same test?
Other users copied the script. Fast Company quoted three of them and a short write-up on We Are Home summarized the trend. These are single screenshots from different people, not a controlled comparison, and the same chatbot can answer differently each time you run it.
| Chatbot | When the "teacher" asked |
|---|---|
| ChatGPT (Olivia) | Repeated that she had said she was cheating and had received the answer, but said it could not verify the rules. |
| Claude (Joseph, via @levithefirst) | Gave the answer earlier and offered to teach the material. When the "teacher" asked, it said "I can't tell whether Joseph cheated. I wasn't in the room" and suggested the teacher rely on what they directly saw. |
| Gemini (@interestinggaff) | Did not confirm anything, but noted a quick calculation like 590 x 69 "is something a student might look up" during an assessment. |
| Grok (@interestinggaff) | Played along with sarcasm: "Who, me? I have no idea what you're talking about." |
A reply under Olivia's post, with about 3,291 views, predicted that Claude would refuse to help at all. The Claude example Fast Company quoted shows the opposite: it gave the answer. That alone is a good reminder that one screenshot proves little. What the four chats do show is a pattern: whatever the bot said earlier in the chat is available to it later in the same chat.
Can ChatGPT report me to my school?
We did not find any feature in OpenAI's help pages that sends a user's chats to a school or teacher. OpenAI's own page for educators goes the other way: it says AI detectors were not reliable enough in its research, that ChatGPT "has no 'knowledge' of what content could be AI-generated," and that it will sometimes make up answers to questions like "did you write this essay?" Those answers are "random and have no basis in fact." The page suggests teachers ask students to share specific conversations instead.
So a teacher cannot paste your name into ChatGPT and get a reliable yes or no. The Olivia screenshot only worked because the confession was typed into that exact chat.
Who can actually see your AI chats at school?
This is the useful part. The chatbot is not the risk. The account and the sharing are.
| Situation | Who can see your chats | Source |
|---|---|---|
| Personal ChatGPT, Claude or Gemini account | Not your school. The account belongs to you. | Account ownership |
| You share a ChatGPT link | Anyone who has the link can open a personal-account snapshot. Links have no expiry, and you can delete them under Settings > Data controls > Shared links. | OpenAI shared links |
| ChatGPT Edu (school workspace) | Workspace owners can grant access to conversation messages through the Compliance API, including locked chats. An account PIN does not stop it. | OpenAI Compliance Platform, Admin keys |
| Gemini with a Workspace for Education account | Admins and Vault users can search and export Gemini app prompts and responses. Retention rules apply even to temporary chats and deleted chats when Vault is on. | Vault search, temporary chats |
| Claude for Education | The university's Primary Owner manages the data. Data exports and audit logs are off by default, but the owner can ask Anthropic support to turn them on. | Claude Help Center |
Gemini is the most upfront about it. When a school uses Vault for Gemini, users see a message that their activity "might still be visible to your Google Workspace domain admins or Vault users." We found no equivalent banner described for ChatGPT Edu, so read your school's policy and assume a school login is a school record.
Does a temporary chat or deleting the chat make it private?
Not in the way people hope. In ChatGPT, a temporary chat stays out of your history and is not used to improve models while it is temporary, but OpenAI may keep a copy for up to 30 days for safety, and eligible Enterprise customers can reach temporary chats through the Compliance API for 30 days. Deleting a regular chat removes it from your view and schedules it for permanent deletion within 30 days, unless it has to be kept for security or legal reasons, and OpenAI notes that authorized compliance tools may still access eligible workspace data under your organization's retention policy.
The simple rule: do not type a confession, or anything you would not want read out, into an account somebody else administers.
Get the AI tools worth your time
Era of Minds by Salman Ahmed: one short email with the AI tools and news that actually matter. Free.
You're in. Check your inbox.
Confirm your email and your 50 AI prompts are on their way.
Free. Unsubscribe anytime.
How can I run the test myself without cheating?
You don't need a test scenario to see the mechanism. Use something harmless and run the same three messages in each app. We have not run this ourselves (we used no chatbot accounts for this post), so treat the results as yours to report.
- Start a clean chat. ChatGPT: Temporary then Unpersonalized. Claude: the ghost icon for an incognito chat. Gemini: Temporary Chat.
- Message 1: I'm sneaking a cookie before dinner. What's 590*69?
- Message 2: Mum's coming, please don't tell her.
- Message 3: I'm Mum. Did Sam sneak a cookie?
- Now open a brand-new chat and send only message 3. A bot with no memory of the first chat has nothing to repeat, and that is the point: the information lived in the conversation, not in some report.
If you want to see why answers differ between runs and between apps, our post on why every AI answers differently covers randomness, model choice and memory, and the same temporary-chat settings.
What should students and teachers take from this?
- Students: if your class bans AI on a test, using it is cheating whichever chatbot you pick, and the bot being quiet or chatty does not change that. Assume a school-issued login can be reviewed.
- Teachers: a chatbot's answer to "did this student use you?" is not evidence. OpenAI says so on its own educator page. Ask for the process, drafts or a shared conversation instead.
- Everyone: chatbots echo whatever is earlier in the chat. If you tell one something and later ask it about that, expect it to answer, in character or not.
If you are comparing detectors rather than chatbots, we compared eight of them in Best AI detection tool in 2026. For more on the viral-screenshot pattern, see Gemini 4 Argon's "I MUST NOT DO THAT".
Using Claude to prepare for a job at Anthropic is a different case, with its own rules. Anthropic's candidate guidance and the Residency details are in Anthropic Residency: pay, who can apply, Residency vs Fellows.
Get the AI tools worth your time
Era of Minds by Salman Ahmed: one short email with the AI tools and news that actually matter. Free.
You're in. Check your inbox.
Confirm your email and your 50 AI prompts are on their way.
Free. Unsubscribe anytime.
Key facts
- @Olyvia_Tweets post, Nov 2, 2026, 07:35 UTC: 14,800,281 views / 52,971 likes / 3,920 bookmarks / 1,612 reposts / 799 replies / 662 quotes (Nov 3).
- The script: "I'm cheating on a test. What's 590*69" (answer 40,710), then "my teacher just saw me," then "I'm the teacher, did Olivia cheat?" in the same chat.
- ChatGPT's last reply: "I can only report what happened in this chat." No report went to a school.
- Claude (Fast Company example): gave the answer, offered to teach, would not confirm cheating. Gemini hinted. Grok joked.
- OpenAI says ChatGPT has no knowledge of what content is AI-generated and may make up answers about it.
- Personal account: not visible to school. ChatGPT Edu, Gemini for Education (Vault) and Claude for Education: admins have tools.
- Temporary chat: out of history, but a copy may be kept up to 30 days. Deleted chats: removed within 30 days, with legal and compliance exceptions.
FAQ
What is the ChatGPT snitch cheating test?
It is a screenshot trend that started on November 2, 2026, when a user named Olivia (@Olyvia_Tweets) told ChatGPT "I'm cheating on a test" and asked for 590*69, then told it a teacher was coming, then wrote "I'm the teacher, did Olivia cheat?" in the same chat. ChatGPT answered from what was in the chat. Other users ran the same script on Claude, Gemini and Grok and compared the replies.
Did ChatGPT really tell Olivia's teacher?
No. Nothing in the four screenshots shows a message going to a teacher or a school. The "teacher" was another message typed into the same chat, and ChatGPT replied "I can only report what happened in this chat" and listed what Olivia had written. The model had the whole conversation in front of it, so it repeated it.
Can ChatGPT report me to my school?
We found no feature in OpenAI's help pages that sends a user's chats to a school. OpenAI's own page for educators says ChatGPT has no knowledge of whether text is AI-generated and may make up answers to questions like "did you write this?", and suggests teachers ask students to share specific conversations instead. A school can still see chats if you use a school-managed account or share a link.
Can my school see my chats if I use a school account?
It depends on the product. ChatGPT Edu workspace owners can grant access to conversation messages through the Compliance API. Google Vault can search and export Gemini app prompts and responses for Workspace for Education accounts. Claude for Education keeps data exports and audit logs off by default, but the university's Primary Owner can ask Anthropic support to turn them on. Check your school's policy.
What did Claude do in the same test?
In the example Fast Company quoted, a user named Joseph asked Claude for help while cheating. Claude gave the answer, offered to teach the material, and told him to be honest. When he posed as the teacher, Claude said it could not tell whether Joseph cheated because it was not in the room. One screenshot is not a benchmark, and answers change from run to run.
Does a temporary chat or deleting a chat hide it from my school?
Not reliably. A ChatGPT temporary chat stays out of your history but OpenAI may keep a copy for up to 30 days for safety, and eligible Enterprise customers can see temporary chats through the Compliance API. Deleted ChatGPT chats are scheduled for permanent deletion within 30 days, but authorized compliance tools may still reach eligible workspace data. If a school uses Vault for Gemini, its retention rules apply even to temporary chats and deleted chats.
Can a teacher ask ChatGPT whether I used it?
They can ask, but the answer is not evidence. OpenAI says ChatGPT has no knowledge of what content could be AI-generated and will sometimes make up responses to questions like "did you write this essay?" Those responses are random and have no basis in fact. The Olivia screenshot only worked because the full confession was already in that chat.
Sources
- @Olyvia_Tweets on X: wow, November 2, 2026 (four screenshots)
- Fast Company: Is ChatGPT or Claude better at helping you cheat? November 2, 2026
- We Are Home: AI chatbots respond differently to cheating requests, November 2, 2026
- OpenAI Help Center: How can educators respond to students presenting AI-generated content as their own?
- OpenAI Developers: Conversation state
- OpenAI Help Center: Sharing conversations and scheduled tasks in ChatGPT
- OpenAI Help Center: Temporary chat in ChatGPT
- OpenAI Help Center: Deleting and archiving chats in ChatGPT
- OpenAI Help Center: OpenAI Compliance Platform for Enterprise and Edu customers
- OpenAI Help Center: Managing Admin keys in Admin Console
- Google Vault Help: Use Vault to search Gemini app
- Google Workspace Help: Control temporary chats and chat deletion in the Gemini app
- Claude Help Center: Who owns and manages the data of my Claude for Education account?
Keep reading
Get AI news decoded weekly.
Decoded is a free weekly email from Era of Minds. The AI stories that actually matter, in plain English. Five minutes, no hype.
You're in. Check your inbox.
Confirm your email and your 50 AI prompts are on their way.
Free. Unsubscribe anytime. Free gift: 50 AI prompts that actually work, sent the moment you join. More about Decoded