With Deepfake Simulations, you test how your employees react to an AI impersonation on a video call. Employees receive a meeting invitation by email and join what looks like a Microsoft Teams call. In the call, a deepfake of a real colleague, for example your CEO, makes a suspicious request and posts a link in the chat. You see how many employees joined the call and selected the link. Employees who select it learn what gave the attack away.
The deepfake is an AI-generated video of a real person in your company who has agreed to it. That person has to give their consent and record a photo and a voice sample on their phone or tablet before you can create the simulation. Plan some time for this, and check early who in your company needs to be involved, for example for data protection or employee representation.
Consent comes first. A deepfake is only created after the person has agreed to it themselves, and only for the scenario they agreed to. They can withdraw their consent at any time. Their deepfake then stops being used immediately, and their photo, voice and deepfake are deleted within 24 hours.
How a Deepfake Simulation works
A Deepfake Simulation is set up like a phishing simulation, with a deepfake scenario instead of a regular phishing template. For employees, it runs in 4 steps:
- Meeting invitation: The employee receives an email that asks them to join a short call. The email contains a link that looks like a Microsoft Teams meeting link.
- The call: If they select the link, a meeting page opens in the browser. It looks like Microsoft Teams, but it isn't. The employee decides whether to join the call.
- The deepfake: In the call, a pre-recorded deepfake video of your colleague plays. The deepfake makes an urgent request and posts a link in the meeting chat.
- Learning moment: If the employee selects the link in the chat, they land on a learning page. It explains that the call was a simulation and which warning signs they could have noticed.
Employees can stop at any point. They can ignore or report the email, not join the call, leave the call or not select the link. Everyone who joins the call receives an email after the video ends that explains that the call was a simulation and that the person in the video was generated with AI.
Data privacy and consent
A deepfake uses the face and voice of a real person, so consent is built into every step:
- You can only create deepfakes of people in your user data, and only after they agree themselves. No one can consent on their behalf.
- The deepfake only appears in the scenario the person agreed to, and it only says what's in the scenario's fixed script.
- Consent is valid for 180 days. The person can withdraw it at any time. Their deepfake then stops being used immediately, and their photo, voice and deepfake are deleted within 24 hours.
For the full details, see Data privacy and consent in Deepfake Simulation.
Scenarios
A scenario defines the story of the simulation: the invitation email, what the deepfake will ask in the video and the link in the chat. SoSafe provides the scenarios. There are currently 2:
| Scenario | What happens | Who to ask | Who it's for |
|---|---|---|---|
| CEO Fraud | A member of the management team asks for an urgent, confidential payment to a new partner and posts a payment approval link in the chat. | Your CEO or CFO | Intended for employees who can approve or release payments, for example in Finance. We recommend sending it to these groups only. |
| IT Help Desk Impersonation | A colleague from the IT help desk asks employees to confirm their login on a linked page because of a security update. | An IT technician or admin | All employees or selected groups. |
Before you start
To run Deepfake Simulations, you need:
- Deepfake Simulation activated for your company. Deepfake Simulations are part of the Advanced package. To have them activated, contact your Customer Success Manager.
- The right admin role. Administrators and Campaign Managers can use the Character Library and create Deepfake Simulations. See Admin user roles and permissions.
- A colleague who agrees to a deepfake of them. They need to be in your user data (they don't need to be an admin user in the SoSafe Manager) and have a phone or tablet to record their photo and voice sample. To add users, see User management: adding and editing individual users.
- Whitelisting. As with phishing simulations, a Deepfake Simulation only starts after you've confirmed that whitelisting is done. The domains for the meeting page and the chat link are listed under Settings / Whitelisting, together with the other simulation domains. See Whitelisting.
- Your data protection setup in place. Your company is responsible for the data used in a deepfake (it's the data controller). Depending on your company and country, you may need to complete a data protection impact assessment (DPIA) or involve your works council first. See Data privacy and consent in Deepfake Simulation.
Get started
- Check the privacy requirements with the people responsible in your company: Data privacy and consent in Deepfake Simulation.
- Get a deepfake of a colleague:
- Talk to the colleague whose deepfake you want to use, then request their consent in the Character Library: Request consent in the Character Library.
- Your colleague gives their consent and records their photo and voice sample on their phone or tablet. You can send them this article to read first: Give consent and create your deepfake.
- Once the deepfake is ready, create the simulation: Create a Deepfake Simulation.
- Follow the results in Analytics: Deepfake Simulation metrics.
What you can measure
Deepfake Simulations have their own channel in Analytics, separate from email and SMS simulations. You see the metrics you know from phishing simulations, such as the click rate, the Interaction rate and the reporting rate. For Deepfake Simulations, the Interaction rate shows how many of the employees who joined the call selected the link in the chat. In addition, the Join meeting rate shows how many of the employees who selected the meeting link joined the call. For details, see Deepfake Simulation metrics.
As with phishing simulations, you choose a reporting method for each simulation. With User group based reporting, you only see results per user group. With Individual reporting, you see how each employee reacted.
Good to know
Is the deepfake a live conversation?
No. The deepfake is a pre-recorded video, so employees can't talk to it. This keeps what the deepfake says under control: it only says what's in the scenario's fixed script, which the person agreed to. No one can get it to say anything else. Every employee also sees the same video, which makes the results comparable.
Does the simulation run in Microsoft Teams?
No. The call takes place on a web page that looks like a Microsoft Teams meeting. Nothing is installed in or sent through your Microsoft Teams environment.
Which languages are available?
A deepfake can speak one of the following 6 languages: English, German, Spanish, French, Dutch or Italian. Each deepfake speaks 1 language: the language set in the user profile of the person it's based on. If that language isn't supported yet, the deepfake speaks English. The deepfake isn't translated into other languages, and dialects such as Swiss German aren't supported.
You can still send a Deepfake Simulation to employees whose language is different from the deepfake's. When you choose the audience, the SoSafe Manager shows which language the scenario is delivered in and how many of your users have that language.
Can I change the script or the emails?
No. The video script, the invitation email and the learning page text are fixed for each scenario.
How many scenarios and deepfakes can I use?
You can have 1 deepfake per scenario: currently 1 for CEO Fraud and 1 for IT Help Desk Impersonation. Each simulation uses 1 scenario. To use both scenarios, create 2 simulations.
Each scenario is delivered in the language of its deepfake. For example, if your CEO's deepfake speaks German, CEO Fraud is delivered in German to everyone who receives it. When you choose whom to ask for a deepfake, think about the language most of the recipients speak.
Does my learning page branding apply?
Partly. The learning page of a Deepfake Simulation doesn't support all branding settings. The following settings aren't applied:
- the legacy color palette (secondary, footer, marker, button and text colors)
- custom text on the learning page
- the feedback and Phishing Report Button settings