At first from April, an influx of action began to appear on social media sites, in particular Liendin And X. Each figure represented the person who had created it with strange precision, with personalized accessories such as reusable coffee cups, yoga mats and headphones.
All this is possible because of the new Openai GPT-4O image generatorwhich overeating chatgpt capacity to modify the images, to make text, etc. Openai's chatgpt image generator can also create images in the style of the Japanese animated film studio Ghibli – a trend that quickly became viralAlso.
The images are fun and easy to do – everything you need is a free chatgpt account and a photo. However, to create an action figure or an image of Ghibli-Studio style, you must also put a lot of data to Openai, which could be used to form its models.
Hidden data
The data you provide when you use an AI image editor is often hidden. Whenever you download an image on Chatgpt, you potentially put back “an entire metadata package”, explains Tom Vazdar, president of the cybersecurity region at the Open Institute of Technology. “This includes exif data attached to the image file, such as the moment when the photo has been taken and the GPS coordinates of their place.”
OPENAI also collects data on the device you use to access the platform. This means that your type of device, the operating system, the browser version and the unique identifiers, explains Vazdar. “And because platforms like Chatgpt work in conversation, there are also behavioral data, like what you have hit, what type of images you asked, how you interacted with the interface and the frequency of these actions.”
It's not just your face. If you download a high resolution photo, you also give OPENAI everything in the image – background, others, things in your room and everything that was readable, such as documents or badges, explains Camden Woollven, AI product marketing group manager at the Risk Management Society GRC.
This type of data supported by the consent provided voluntarily is “a gold mine for the formation of generative models”, in particular the multimodal models which are based on visual entries, explains Vazdar.
Openai denies that he orchestrated trends in viral photo as a ploy to collect user data, but the company certainly draws an advantage. Openai does not need to scratch the web For your face, if you download it with pleasure yourself, underlines Vazdar. “This trend, whether by design or a practical opportunity, provides the company with massive volumes of fresh and high quality facial data from various age groups, ethnic groups and geographies.”
OPENAI says he is not actively looking for personal information for train models– And it does not use public data on the Internet to create profiles on people to advertise it or sell their data, an Openai spokesperson told Wired. However, under the current of Openai Privacy Policyimages submitted via chatgpt can be kept and used Improve your models.
All the data, prompts or requests that you share help to teach algorithm – and personalized information helps to refine it more, explains Jake Moore, World Advisor to Cybersecurity of Safety ESET, which has created its own figure of action to demonstrate the risk of confidentiality of the trend is Linkedin.
Strange resemblance
In some markets, your photos are protected by regulations. In the United Kingdom and in the EU, the regulations on data protection, including GDPR Offer solid protections, including the right to access or delete your data. At the same time, the use of biometric data requires explicit consent.
However, photographs become biometric data only when processed by a specific technical means allowing unique identification From a specific individual, explains Melissa Hall, main partner of the MFMAC law firm. The processing of an image to create a cartoon version of the subject on original photography is “unlikely to respect this definition,” she says.
