I read a few month ago a comment from an Anthropic or OpenAI employee saying that the security problem of prompt injection was a solved problem at this point, as the frontier models where rejecting 98% of attacks.

It tells you all you need to know about how unserious these people are about computer security.