AI Risk Management: Deepfakes

‘Deepfake’ technology is all about faking content, and is becoming more sophisticated and accessible to malicious actors. Generally, these are created using a deep learning model called GAN (generative adversarial network), which are trained on both the source data and the fake data (target set). These are combined to create a deepfake that cannot be distinguished from the real image.

As AI technology becomes easier to use, this risk/problem is getting worse.

Key risks include

• damaging peoples reputations
• blackmail and extortion
• spread of misinformation
• national security

On the flip side, deepfakes are useful for creating realistic special effects for TV/Movies.

How do we mitigate these risks?

• Check the authenticity of all information
• Be critical of anything you read or see online
• Use tools or resources to detect deepfakes