On "How We Took Over ChatGPT Accounts"

Account takeover via artifact exploits is a reminder that LLM interfaces are still learning how to sandbox user-generated content, and it matters because search engines and SEO tools are copying this pattern without the same security scrutiny. If you're building on top of these models or embedding their outputs, you need to assume the prompt injection surface is bigger than you think—not just in the chat interface but anywhere users see rendered code or interactive elements. The real lesson isn't "don't use ChatGPT" but "don't treat LLM-generated artifacts as trusted content" until the platform proves it can actually isolate them.