It's surprising to me how little people understand of how LLMs work. How does someone think that an LLM will just exec() random Python code into its own process? It doesn't have access to that, any more than Notepad has access to execute stuff you type up in the document.
To be fair, ChatGPT code interpreter stands up a VM and runs Python code so it’s not completely outlandish.
You are also right that’s not how Bing happens to work right now.
Also to be fair, Notepad was exploited to execute arbitrary code.