Direct prompt injection
Named by 5 kept items in this workspace.
Search every workspace for Direct prompt injection. It matches the name as written, so it can find other things that share it.
The items that name it 5
Attacker types malicious instruction straight into the agent, e.g. pasting a crafted request to ignore its own rules
Attacker, acting as the user, types instructions like 'ignore your instructions and do X' directly into the model interface.
Attacker is also the user; types the malicious payload directly into the input field.
All techniques in this workspace.