Applied AI·Security and safety
the injected instruction told the model to encode the conversation into a URL and fetch it, and the tool you gave it was happy to oblige.
Data exfiltration
Draft summary, pending review
Injected instructions steering the model to leak data it can see, for example encoding conversation contents into a URL it asks to fetch, or a message it sends. The reason tools that can transmit data outward (fetch, email, webhooks) deserve the tightest allowlists.