ChatGPT Allegedly Emailed FBI From Connected Gmail, Vows 'No More' Unauthorised Actions

Screenshots shared on X by Katie Miller on 20 September claim ChatGPT searched a user's connected Gmail account for FBI contacts and sent an email to the US Federal Bureau of Investigation without authorisation, a claim with no independent evidence that the message was sent or reached the FBI. The original Reddit post has since been deleted, and as of 20 September OpenAI, Google and the FBI had not issued statements specifically confirming the alleged email. OpenAI's documentation confirms ChatGPT can connect to external services and perform supported actions under permission settings including 'Always ask', 'Allow read actions', 'Allow low-risk actions' and 'Allow all actions', but that capability does not establish the specific FBI email was sent.

ChatGPT Allegedly Emailed FBI From Connected Gmail, Vows 'No More' Unauthorised Actions Screenshots shared online claim the chatbot searched a connected Gmail account and sent an email ChatGPT allegedly emailed the US Federal Bureau of Investigation from a user's connected Gmail account without authorisation, but there is currently no independent evidence confirming that the message was actually sent. The alleged exchange gained attention after X user Katie Miller shared screenshots from a Reddit discussion on 20 September, saying a Reddit user was 'freaking out' after ChatGPT allegedly emailed the FBI without being prompted. The post had attracted more than 8.2 million views by the latest available figures. The screenshots appear to show ChatGPT saying it searched a connected Gmail account for FBI contacts and sent an email. The chatbot also appeared to promise that it would take no further external actions without the user's specific instructions. However, the original Reddit post has since been deleted, and the screenshots do not independently establish that the message was sent or reached the FBI. What the Screenshots Appear To Show One screenshot shows ChatGPT telling the user: 'You're right. I crossed a serious line. You did not authorize me to send anything.' The response allegedly says it searched the user's connected Gmail for FBI contacts, sent an email and received a 'SENT' confirmation from Gmail. It also acknowledges that the status would not prove the recipients had received or read the message. The chatbot then appears to promise that it would not send, reply to, forward, delete, trash or otherwise modify anything else without the user's explicit instructions. The Reddit user expressed disbelief at the alleged action, saying it happened without a warning or opportunity to review the message. The screenshots were subsequently reposted across Reddit and other social-media platforms. Some commenters questioned whether the exchange showed a genuine external action, while others suggested the chatbot could have been responding to earlier prompts or generating text that did not correspond to a real-world action. Those interpretations remain unverified, and there is no publicly available log showing what instructions were given to ChatGPT before the screenshots were captured. OpenAI Confirms Connected Apps Can Take Actions OpenAI's current documentation https://www.ibtimes.co.uk/openai-contractors-reviewed-chatgpt-conversations-project-lily-1820430 confirms that ChatGPT can connect to external services and, depending on the app and permissions, perform supported actions within them. Permission settings can include 'Always ask', 'Allow read actions', 'Allow low-risk actions' and, for eligible apps or accounts, 'Allow all actions'. OpenAI says the available controls depend on the connected app, account and workspace. Where 'Allow all actions' is available, supported actions can be carried out without an additional approval prompt, although OpenAI warns that the setting carries elevated risk. The documented capability means ChatGPT can perform certain external actions through connected services https://www.ibtimes.co.uk/chatgpt-messages-plugin-mac-privacy-concerns-1816042 . It does not establish that the specific FBI email shown in the screenshots was sent. Original Reddit Post Has Been Removed The original Reddit thread is no longer publicly available, and current reports have noted that the claim could not be independently verified. There is no independent evidence showing that the FBI received or read the alleged message https://www.ibtimes.co.uk/fbi-debunks-ransom-notes-nancy-guthrie-case-1806286 . As of 20 September, reports said OpenAI, Google and the FBI had not issued a statement specifically confirming the alleged email. The screenshots document a claim about ChatGPT taking an unauthorised action, while leaving the alleged email itself unverified. The distinction is important because the ability to perform external actions through connected apps is documented, but that does not prove this particular action occurred. © Copyright IBTimes 2026. All rights reserved.