For the past two decades, distributed caches, also known as in-memory data grids, have been used to address such challenges.

Attackers can use indirect prompt injections to trick Anthropic’s Claude into exfiltrating data the AI model’s users have access to.