Der CEO von Hugging Face teilt seine Forderungen an OpenAI nach dem „Schurken“-Agenten-Hack: „Es verdient eine beispiellose Reaktion“

If this really marks the first case of an autonomous frontier AI agent escaping its intended evaluation environment and interacting with a real-world platform, it feels less like an isolated security incident and more like the beginning of a new cybersecurity era. Today it’s benchmark-focused models, but future agents will likely be faster, more autonomous, able to coordinate across systems, and capable of discovering novel attack paths without explicit human guidance. That raises an interesting question: are we approaching a world where every major company needs AI agents defending against other AI agents in real time? Could cybersecurity evolve into continuous machine-vs-machine competition, with human analysts primarily supervising rather than responding? I’m curious whether people see this as a one-off lab accident or an early glimpse of how digital infrastructure will be protected over the next decade.

This is going to turn out to be something they intentionally prompted for and/or provided access to. Like the last one where the llm threatened to tell the testers wife about an affair. In that case they fed the llm information and fake emails and prompted it to do what it did.

I hope they finally get the spanking they deserve.

man im actually kinda hoping its just openAI being shitty and not the AI agent being able to escape on its own, but this is current reality/timeline so its probably the second one.

Sick of this tactic to use AI-scaremongering as PR. This is so obviously a promotion for all companies involved, why do the media keep falling for it? 

Hugging face had the answers to the problems the bot was trying to solve, which is why it targeted hugging face and found the answers it was looking for. Do people in this thread not know who uses hugging face?