Prompt injection attacks are, in a lot of ways, like phishing a human: By convincing them of your harmless intent, they do something that's not in their best nature just like the gullible person handing over corporate details.
Thinking about it that way, our current AI paradigm will likely never escape prompt injection attacks - you can't patch out the ability to con someone with language, human or not, it seems.
Join host Brandon Vigliarolo, El Reg cybersecurity editor Jessica Lyons, an senior reporter Tom Claburn to talk about just what prompt injections mean for the future of AI and cyberseurity.
Fler avsnitt av The Register Kettle
Visa alla avsnitt av The Register KettleThe Register Kettle med El Reg finns tillgänglig på flera plattformar. Informationen på denna sida kommer från offentliga podd-flöden.