3 ms·
It's a security problem if you care about keeping your prompt secret. It's even more of a security problem if you plan to plug your language model into somethi
by simonw 4y ago
It's a security problem if you care about keeping your prompt secret.
It's even more of a security problem if you plan to plug your language model into something that can execute additional actions. People have already been caught out running generated code through eval() - and there are plenty of potential applications for things like customer support bots that cancel accounts or offer discounts.
Developers who are unaware of prompt injection are very likely to make dangerous design mistakes!