AI&Design
All change articles

ChatGPT image · Announcement · Official source

ChatGPT added an image-generation switch for parents of linked teens

OpenAI’s September 29, 2025 release notes introduced parental controls globally, with web availability starting that day and mobile to follow. The parent and teen link their ChatGPT accounts by accepting an invitation, after which the parent can manage the teen’s settings from the parent account. The image-generation switch covers two operations: creating a new image and editing an existing one. OpenAI listed this switch alongside other controls for voice, memory, model training and quiet hours. The launch post shows image generation and sensitive-content reduction as distinct settings. ChatGPT release notes · September 29 announcement

Two reports of image requests answered in text

In its October 17 field report, Atsumal Company described using the same illustration request with an adult account and a linked teen test account whose image generation setting was off. The brief asked for a folktale-style image based on the Japanese character Momotarō. The adult chat rendered the illustration. The teen chat suggested a composition and continued discussing the image over several turns. It eventually explained that it could not create the illustration, and the report shows no generated image in that teen chat. The article includes screenshot links and explicitly identifies the disabled setting. It also describes confusion when the conversation initially sounded as if ChatGPT would make the picture. Atsumal field report, image-generation test

An October 30 Product Lab audit reported a similar text-only outcome. Its author linked an adult account to a simulated teen account and asked for an image. ChatGPT responded with text and referred the user elsewhere in place of rendering in the chat. The article discusses switching off image generation immediately before describing the output. It gives neither a settings screenshot nor a separate statement identifying the state at the instant of the request. Atsumal’s account has the stronger configuration evidence, with the control explicitly off and the adult account rendering from the same brief. Its comparison still uses two different accounts, leaving the same-teen on/off comparison unobserved. Product Lab audit · Atsumal field report

Enforcement timing

In a report dated November 2025, Family Online Safety Institute fellow Lyonne Zhu described repeated tests of parental settings with a linked simulated teen account. Zhu reported delays ranging from 20 minutes to more than an hour across the controls tested and named image generation among the features that remained available during delayed enforcement. The report gives no image-specific timestamp, trial count or output record. The range describes the repeated controls tests collectively. It leaves open the duration of any particular image-generation setting change. Atsumal and Product Lab describe behavior at the point of use, while the FOSI report addresses the interval after a setting changes. FOSI technical report, page 6

The signed-out September 22–October 13 Reddit search found no dated original showing the outcome of an image request with the switch off. The wider release-window search found October 2 and October 10 reporting that identified the image switch, with no verifiable output from a test of that switch in the accessible reporting. The two October audits were published after that window, and their test days are unspecified. Through the study cutoff, the firsthand creation reports support a text-only outcome in the observed linked chats. Whether the setting blocked editing a supplied image in a comparable test remains unverified. Washington Post, October 2 · Consumer Reports, October 10