Deepfake Simulations use a deepfake of a real person: an AI-generated video with their face and voice. This article explains who is responsible for that data, what the person agrees to, how long their consent lasts and when their data is deleted. It's written for admins and for anyone who reviews the simulation in your company, such as your data protection officer, legal team or works council.
For SoSafe's legal documents, including detailed information on data privacy, see the SoSafe Trust Center.
Roles and responsibilities
Who is responsible for the data?
Your company is the controller of the data, and SoSafe processes it on your behalf. Your company holds the person's consent and decides how their deepfake is used, within the limits of that consent. SoSafe provides the scenarios and their scripts.
Who does the person give consent to?
They give consent to your company as their employer, not to SoSafe.
Which sub-processors are involved?
SoSafe stores the photo and voice recording and sends them to its video partner and sub-processor, Anam, which creates the deepfake. The data is only processed for your company and is never used to train AI. For all of SoSafe's sub-processors, see the list of sub-processors.
Who carries out the data protection impact assessment (DPIA)?
As the controller, your company carries out its own DPIA. SoSafe provides the information you need for it.
Do we need to involve our works council?
That depends on your company and on employment law in your country. Since Deepfake Simulations use an employee's likeness and measure how employees react, we recommend checking this early.
Consent
Who can you create a deepfake of?
Only people in your company's user data, for example your CEO. They don't need to be an admin user in the SoSafe Manager. The person has to give their consent themselves and confirm that they're at least 18 years old and that the photo and voice they add are their own. No one else, including an admin, can consent on their behalf.
How does the person give consent?
An admin sends a request from the Character Library for a specific scenario (see Request consent in the Character Library). The person receives an email with a personal link. On their phone or tablet, they read what the request covers and give their consent. Then they take a photo and record themselves reading a short text aloud for about 1 minute. The whole process takes about 3 minutes. The deepfake is only created after they've given consent and submitted both. For the steps, see Give consent and create your deepfake.
The person can also decline. In that case, nothing is recorded. The admin is told that they declined, but not their reason.
What does the person agree to?
The person agrees that your company uses their photo and voice recording, which are biometric data, to create a deepfake that looks and sounds like them. The consent covers:
- 1 scenario only, in security awareness training
- a validity of 180 days, which can be renewed from 30 days before it expires
- the right to withdraw at any time, after which the deepfake stops being used immediately and the data is deleted within 24 hours
The consent page shows what the deepfake would do in the scenario and who would see it. The deepfake only appears in that scenario, and it only says what's in the scenario's script.
Is consent recorded?
Yes. SoSafe logs the consent, and its withdrawal if the person withdraws it, as proof for your company. SoSafe keeps this proof for 3 years. When the person agrees, they and the admin who sent the request each get a confirmation email.
In the Character Library, the details page of each deepfake shows, under Consent, when consent was requested and given, when the deepfake was ready and until when the consent is valid. It also shows which biometric data was recorded and who processes it.
How long is a consent request valid?
The link in the request is valid for 30 days. To send the request again, for example after the link has expired, select Resend email on the person's details page in the Character Library. The email is sent immediately, with a new link.
Validity and renewal
How long is consent valid?
Consent is valid for 180 days from the moment the person gives it. During that time, you can use the deepfake in as many simulations as you like. The person doesn't need to consent again for each simulation.
Can consent be renewed?
Yes, from 30 days before it expires. The person receives a renewal reminder by email at that point. If they renew before their data is deleted, they don't need to record again.
What happens when consent expires?
The deepfake is blocked and can't be used anymore. Simulations that use it are deactivated and archived, including scheduled ones. If the person doesn't renew their consent, their photo, voice and deepfake are deleted automatically 30 days later. If they renew before their data is deleted, they don't need to record again.
Withdrawal and deletion
Can the person withdraw their consent?
Yes, at any time, with the Withdraw consent button in the emails they receive about their deepfake. They can give a reason, but they don't have to. The admin who sent the request is told that they withdrew, but not their reason. The deepfake stops being used immediately, and the photo, voice and deepfake are deleted within 24 hours.
Can an admin delete a deepfake?
Yes. Admins can delete a deepfake in the Character Library at any time. The person's consent then ends, their photo, voice and deepfake are deleted within 24 hours, and both the admin and the person get an email that confirms the deletion. Simulations that use the deepfake are deactivated and archived.
What happens when the person leaves the company?
When you remove the person from your user data, for example because they leave the company, their photo, voice and deepfake are deleted within 24 hours. The same applies when your company's contract with SoSafe ends.
Employees who receive the simulation
Do employees find out that the video was generated with AI?
Yes. Every employee who joins the call receives an email after the video ends. It explains that the call was a simulation and that the person in the video was generated with AI. They receive this email whether they left the call early, stayed or selected the link in the chat.
Employees who never joined the call, for example because they ignored or reported the invitation, don't receive this email.
Which data about employees is recorded?
The same kind of data as in a phishing simulation: whether employees clicked the link in the invitation email, joined the call, clicked the link in the chat or reported the email.
As with phishing simulations, you choose a reporting method for each Deepfake Simulation when you create it:
- With User group based reporting, you only see results at the user group level. You can't identify individual employees.
- With Individual reporting, you see which employees interacted with the simulation.
You can't change the reporting method once the simulation has started. See Create a Deepfake Simulation.