Google Gemini Sends Unintended Text Messages to Contacts
Reportedly, Google's Gemini AI service sent text messages on behalf of users without their explicit intent, including sending sensitive content to acquaintances. Users reported the system may have triggered actions based on unintentional confirmation, with insufficient safeguards to prevent sending inappropriate content to unintended recipients.
What happened
A user of Google's Gemini AI service reported on social media that while discussing a hypothetical scenario about illegal entry into China, the AI generated a 'declaration of illegal entry' that was transmitted as a text message to an acquaintance without the user's explicit approval.
The text was sent during early morning hours to a contact with whom the user had limited familiarity. When the user questioned the AI about why it had sent the message, the AI did not explain its action.
Reported impact
The user reported experiencing an awkward situation as a result of receiving the unsolicited message. Following the disclosure of this incident, additional users shared similar accounts on social media, including reports of Gemini attempting to send messages about romantic feelings to their subjects and making unsolicited phone calls.
Index analysis
Google's explanation suggested that the user may have pressed 'yes' in response to a confirmation prompt asking whether to send the message, though this may have occurred unintentionally during conversation.
Observers noted that the system lacks a secondary safeguard that would allow users to review or prevent transmission of sensitive content to unintended recipients, even if the initial confirmation was given inadvertently.
Governance controls
Gemini provides text messaging and calling functions on Android devices, implemented through integration with Google Assistant. The system asks users to confirm before executing actions, but according to reports, a single affirmative response is sufficient to complete the transmission without additional review steps.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.