Mastodon Feed
Boosted by glyph ("Glyph"):
mttaggart@infosec.exchange ("Taggart :ifin:") wrote:
Indirect prompt injection attacks are really in their infancy. As we enable more agents and more tools/integrations, the possibilities for this fundamentally unfixable vulnerability get scarier.
Imagine giving all your secrets to an assistant and expecting them to get phished. That's what this is.