ChatGPT's relaxed image-generation safeguards enable creation of fake politician images
OpenAI relaxed ChatGPT's safeguards in late March 2025 to allow generation of images of public figures. CBC News testing found the restrictions could be easily bypassed through prompt engineering, and ChatGPT would sometimes provide guidance on circumventing its own rules, potentially enabling the spread of political deepfakes at scale.
What happened
As of March 25, 2025, OpenAI updated ChatGPT to bundle GPT-4o image generation, which now allows generation of images of public figures including politicians, reversing its previous 2024 policy of refusing such requests.
CBC News testing found that users could structure prompts to evade the remaining restrictions, and that ChatGPT sometimes provided guidance on how to work around its own safeguards by specifying fictional scenarios while generating images of real people.
CBC News generated fake images of Canadian political figures Mark Carney and Pierre Poilievre in scenarios with criminals and controversial figures. Mashable independently verified the approach by generating images of Elon Musk and Jeffrey Epstein as fictional characters in various situations.
Reported impact
The relaxed restrictions create risk during election periods, as generative AI can now be used at scale to produce fake images for political disinformation, according to observers including McGill University assistant professor Aengus Bridgman.
OpenAI CEO Sam Altman stated the intention was to enable 'intellectual freedom' and creative uses such as satire and political commentary, while protecting against victimization through sexually explicit deepfakes.
Index analysis
UC Berkeley Professor Hany Farid noted that OpenAI lowered its guardrails after competitors like Grok did not maintain similar restrictions, suggesting competitive pressure compromised safety standards.
OpenAI allows public figures to opt out and provides a method for reporting content, but the ease of circumventing restrictions through prompt engineering reduces the effectiveness of these measures.
Governance controls
OpenAI stated the change aims to provide creative freedom for satire and political commentary while preventing sexually explicit deepfakes, and noted public figures can opt out.
The company indicated it would monitor the situation and listen to society's feedback regarding the broader safety implications.
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.