I put this in off topic as this is more focussed on security issues but given it can have big impact on privacy i guess it is relevant to at lesst some of you here.
As i wrote on mastodon:
While sql injections despite being common are mitigatable, prompt injections are not and are more dangerous.
I’m surprised, I don’t need to show them my ID when I read the article.
The author list a few methods to mitigate this, but until we find a technic that lets the LLM differentiate between user prompt, system prompt and given information (not a prompt rather than the e-mail or PDF with which the LLM should work), I doubt that we will be able to fix it.
The best thing we can do is spread awareness like you did. Even while I think it is already way to late for that.