How Indirect Prompt Injections Exploit Context, Format, and Salience (by afogel) — discussion

#ai #security