🕶 OpenAI's Viral Privacy Trick
AI's Legal and Ethical Challenges | Edition #184
Hi, Luiza Jarovsky here. Welcome to our 184th edition, read by 56,800+ subscribers worldwide. It's great to have you here! Paid subscribers have full access to my timely analyses of AI's legal and ethical challenges. Don’t miss out!
For more: AI Governance Training | AI Book Club | Live Talks | Job Board
🕶 OpenAI's Viral Privacy Trick
Last week, I wrote about the Ghibli Effect and how it represents a decisive moment in the AI Copyright debate. If you haven't read it yet, I recommend you read it first to understand what's happening and what the Ghibli effect is.
Many haven't noticed, but the Ghibli Effect also has significant privacy implications. Why?
To get their own Ghibli (or Sesame Street) version, thousands of people are now voluntarily uploading their faces and personal photos to ChatGPT. As a result, OpenAI is gaining free and easy access to many thousands of new faces to train its AI models.
Some people will argue that this is irrelevant because OpenAI could simply scrape the same images from the internet and use them to train its AI models. This is not true, for two reasons:
1. Privacy “Bypass”
In places like the EU, when OpenAI scrapes personal images from the internet, it relies on legitimate interest as a lawful ground to process personal data (Article 6.1.f of the GDPR).
As such, it cannot harm people or go against their interests, and therefore, it must take additional protective measures, including potentially refraining from training its models with these images (see my previous articles on the topic, including on Opinion 28/2024). Other data protection laws specify additional protections in the case of scraped images, including for images of minors.
However, when people voluntarily upload these images, they give their consent to OpenAI to process them (Article 6.1.a of the GDPR). This is a different legal ground that gives more freedom to OpenAI, and the legitimate interest balancing test no longer applies.
Moreover, OpenAI's privacy policy explicitly states that the company collects personal data input by users to train its AI models when users haven't opted out (*you can opt out here).
2. Fresh New Images
My second argument for why this was a clever privacy trick is that people are uploading new images, including family photos, intimate pictures, and images that likely weren't on social media before, just to feel part of the viral trend.
OpenAI is gaining free and easy access to these images, and only they will have the originals. Social media platforms and other AI companies will only see the “Ghiblified” version.
Moreover, the trend is ongoing, and people are learning that when they want a fun avatar of themselves, they can simply upload their pictures to ChatGPT. They no longer need third-party providers for that.
*
OpenAI obtained these new images voluntarily using a simple PR trick. What trick?