AI Agent Security (Course)
Toxic Flow
Also: toxic agent flow
A harmful sequence made of individually harmless tool calls, such as reading private data and then calling an outbound tool.
In more detail
Each step in a toxic flow looks acceptable on its own: reading a file, fetching a page, posting a message. The danger appears only in the combination, typically untrusted input, access to private data and a way to send data out in the same agent session. Because the problem is emergent rather than textual, scanning tool descriptions rarely finds it. The practical defenses are to separate those capabilities across agents or sessions and to watch sequences of calls, not single calls.