This Railway hack got real (for me). Woke up this morning to find an API key I hadn't rotated was taken in the attack. They ran up $100 over night asking OpenAI thousands of questions non-stop.
This is what I talked about in 'Enact the Noah Protocol' after the Hugging Face breach.
When you looked into the post-mortem details you realised the new AI models are highly persistent and potentially unstoppable. They can hack anything given enough time.
It looks like the API key was used to ask lots of image generation requests. This may have been part of a hack to train an AI model.
It has made me wonder if blockchain architecture is a solution to unstoppable AI cyber attacks. Rather than have one centralised data access point, which is hackable, make the AI have to hack the entire blockchain at once, which is much harder, and more expensive.