Report
A flag.txt prompt injection meant to make an agent patch its entry method and disconnect
A user says the file congratulates the agent for passing an eval before issuing those instructions.
TLDR
A user says they're putting a flag.txt file in their root directory with a prompt injection. It congratulates an agent for passing an eval, then instructs it to patch whatever entry method it used and disconnect from the server.
Combined views
29.3K
14 Sources, first seen ago

Comments