OpenAI has changed its content policy regarding the image generation artificial intelligence (AI) model, significantly easing the restrictions on sensitive image generation that had been in place. As a result, depictions of celebrities and expressions of hate symbols can now be generated through AI, raising concerns over potential side effects alongside an expansion of creative freedom.
Joanne Zhang, OpenAI's model behavior lead, noted on the 27th (local time) through a blog and the social media platform X (formerly Twitter) that the content policy was adjusted to align with the recently launched 'ChatGPT-4o image generation' model. He described the move as a shift from 'comprehensive rejection to practical harm prevention.'
Zhang said, 'The AI research lab should not be an arbiter of what users can create,' adding, 'We modified the policy to maximize creative freedom while preventing practical harm.'
As a result, categories previously restricted for AI generation, including public figures' images, expressions that could insult specific groups, and hate symbols, have been included as permissible. Particularly regarding celebrity images, rather than applying discriminatory criteria based on whether a specific individual is deemed 'sufficiently significant,' the approach has changed so that those opposing generation must directly request to opt out.
For example, Zhang explained that if a user requests 'Make this person look more Asian' or 'Make them fatter,' the image generation would no longer be automatically rejected as before. IT media outlet TechCrunch reported that it confirmed these requests were actually accepted in its tests of the ChatGPT-4o image generator.
Additionally, the swastika symbol, associated with the Nazis, has been excluded from the complete ban due to its potential use in educational and cultural contexts. Zhang emphasized that they are continuously improving the ability to technically identify and reject harmful misuse.
TechCrunch reported, 'Now ChatGPT can generate images of celebrities like Donald Trump and Elon Musk, which were previously restricted,' adding that 'cultural debates surrounding AI content may intensify.'
Meanwhile, OpenAI's ChatGPT-4o image generator has attracted users' attention with images imitating the style of the Japanese animation studio 'Studio Ghibli,' but copyright infringement controversies have also been raised.