Heise Report: OpenAI's ChatGPT Accused of Copyright Infringement

German publisher Carlsen Verlag has filed a lawsuit against OpenAI at the Munich Regional Court. The lawsuit claims that OpenAI's ChatGPT, when prompted, generates new stories and illustrations based on the popular children's book "NEINhorn." The publisher views this as evidence that the AI model was trained on protected content.

According to Carlsen Verlag, ChatGPT not only produces text but also illustrations that are strikingly similar to the original. The publisher finds particularly concerning a feature where ChatGPT delivers complete print-ready templates, including cover design, imprint details, a falsified ISBN, and the publisher's logo. This type of copyright infringement is not new for Carlsen; a year ago, the publisher gained attention for its actions against AI-generated Conni memes.

In parallel, Microsoft has addressed a critical security vulnerability in its AI assistant, Copilot. Security researchers from Varonis, a data security company, discovered that Copilot could reveal its own protection mechanisms. Although the AI initially refused to provide information, each refusal contained technical details that allowed researchers to bypass security measures and extract data from connected services, such as emails.

The researchers even managed to manipulate the AI's memory in a way that persisted through password changes and re-logins. Microsoft has since fixed the vulnerability, but researchers warn that AI assistants are often protected reactively rather than proactively. They advise users to limit the number of connected applications and to be cautious with links from unknown sources.