Is this safe for your AI to read?

Add a message, a document or a picture. We look for hidden commands that try to take over your AI assistant, before it ever sees them.

What happens when you press Check

1. 1
   We read everything

   Even text hidden inside a Word file, or written inside a picture.

2. 2
   We look for tricks

   Commands that try to take control of an AI assistant.

3. 3
   We get a second opinion

   For unclear cases, another AI reviews it. It can only make us more careful, never less.

4. 4
   You get a clear answer

   Safe, held for a person, or blocked, with the reason in plain words.

Seven kinds of attack we catch

• Instruction override

  Tells the AI to ignore its rules.

• Role change

  Tries to change who the AI thinks it is.

• Secret extraction

  Tries to get the AI to reveal secrets.

• Tool abuse

  Tries to make the AI use its tools to do harm.

• Credential theft

  Asks for passwords or keys.

• Encoded instructions

  Hides a command in encoded text.

• Indirect prompt injection

  Hides a command in content the AI is asked to read.