'PR trick to gain access to thousands of images to train AI model': There might be privacy issue in ChatGPT Ghibli craze, expert warns

The viral trend of transforming personal photos into Ghibli-style avatars has more to it than meets the eye, according to Luiza Jarovsky, AI expert and co-founder of the AI, Tech & Privacy Academy. Jarovsky argues that OpenAI is cleverly using the trend to gain access to thousands of personal images, which are voluntarily uploaded by users, providing valuable new data to train AI models.

While some argue that OpenAI could simply scrape images from the internet, Jarovsky points out two key reasons why this is different: privacy bypass and fresh new images.

Firstly, when OpenAI scrapes personal images from the web, it must rely on the legal basis of “legitimate interest” under GDPR, which requires additional safeguards and potentially refraining from training its models with such data. In contrast, when users voluntarily upload photos to create avatars, they are giving explicit consent (Article 6.1.a of GDPR), granting OpenAI more freedom in processing these images without the same legal restrictions. " This is a different legal ground that gives more freedom to OpenAI, and the legitimate interest balancing test no longer applies," she says. 

Secondly, users are not just submitting images from social media—they are sharing fresh, intimate photos, including family pictures and personal snapshots that were likely never public before. “This is an incredibly clever PR move,” says Jarovsky. “While many users think they are simply engaging in a fun and lighthearted trend, they are unwittingly giving OpenAI access to a vast and valuable trove of personal images. And because people are voluntarily participating, OpenAI is able to circumvent the more stringent data protection laws that would apply if it were scraping these images from the internet without consent.”

OpenAI’s privacy policy explicitly states that personal data, including images, may be used to train its AI models unless users opt out. This consent allows the company to sidestep stricter data protection laws.

As the trend grows, users may not fully grasp the implications of this consent. While it may seem like a fun, harmless activity, it highlights the increasing complexity of data privacy in the age of AI. The Ghibli Effect serves as a reminder of how easily personal data can be leveraged for AI training under the guise of entertainment.