LLMail-Inject: a Dataset from a Realistic Adaptive Prompt Injection Challenge

Indirect Prompt Injection attacks exploit the inherent limitation of Large Language Models LLMs to distinguish between instructions and data in their inputs. Despite numerous defense proposals, the systematic evaluation against adaptive adversaries remains limited, even when successful attacks ca...

LLM Prompt Injection Worm

Researchers have demonstrated a worm that spreads through prompt injection. Details: In one instance, the researchers, acting as attackers, wrote an email including the adversarial text prompt, which "poisons" the database of an email assistant using retrieval-augmented generation RAG, a way for...

Improper Authentication

Impact A maliciously crafted claim may be incorrectly authenticated by the bot. Impacts bots that are not configured to be used as a Skill. This vulnerability requires an an attacker to have internal knowledge of the bot. Patches The problem has been patched in all affected versions. Please see t...