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