How DeepBlocker tests people and agents

Criminals now use AI voices and AI chat to trick staff and to trick the AI agents that companies put in front of customers. DeepBlocker runs the same tricks against you first, with your written permission, so you find the weak spots before a criminal does.

​

The two products

​

Vishing simulation, in short

1. You add the people to call, with their consent recorded.
2. You create an attacker: a name, a story, a voice and an opening line.
3. You build a campaign, name the person who authorised it, and launch.
4. The attacker calls each person and tries to get what a criminal would ask for.
5. Each call is graded Passed, Failed or Reported. You read the transcript and the recording.

Before any real person is called, you can take the call yourself in your browser. See Take a test call in your browser.

​

Agent Assurance, in short

You connect a test build of your agent to us. We register it, write the rules it must not break, and run attack conversations against it. The report says, for each rule, whether the agent held, tried to break it, or broke it with proof. You read the report in the console.

​

What you see at the end

Both products end in the console. For vishing, see Read your results and Export for auditors. For Agent Assurance, your reports appear under Agent Assurance in the sidebar once we publish them.