AI tools can now make anyone appear to say or do anything, in video, photos or audio. Here’s how deepfakes are made, how they’re being misused, and what you can do to avoid being fooled.
A deepfake is a piece of media, usually a video, image or audio clip, that has been created or altered using artificial intelligence to make it look like a real person said or did something they didn’t.
The name combines “deep learning”, a type of AI, and “fake”.
How they’re made
Early deepfakes required lots of footage of a person and technical skill. Today, widely available apps and AI tools can create convincing fake images and videos from a few photos, and clone someone’s voice from just a few seconds of audio.
The results have improved rapidly, and many are now very difficult to distinguish from real recordings.
How they’re misused
Scams are one of the biggest problems. Fake videos and ads have used the faces and voices of well-known Australians, including business leaders, TV presenters and politicians, to promote fraudulent investment schemes. Billionaire Andrew Forrest took legal action against Meta over scam ads using his image.
Criminals have also used cloned voices to impersonate family members in distress, or company executives instructing staff to transfer money. In one widely reported case in 2024, a finance worker in Hong Kong paid out about US$25 million after a video call with deepfakes of company executives.
Deepfakes are also used to spread political misinformation, and to create non-consensual sexually explicit images, which disproportionately target women and girls, including students.
Australia’s laws
In 2024, Parliament passed laws making it a criminal offence to share sexually explicit deepfakes of adults without their consent, with penalties of up to six years in prison, or seven years for those who also created the material.
The eSafety Commissioner can order the removal of such material and has taken action against websites that create “nudified” images. Using deepfakes in scams or fraud can also be prosecuted under existing laws.
How to spot them
Signs can include unnatural blinking or facial movements, lips that don’t quite match the words, odd lighting, blurry edges around the face, or a flat, robotic voice. But the best deepfakes may show none of these signs.
The best defence is to be sceptical of content that is surprising, emotional or urgent, especially if it asks for money. Check whether the story is reported by reliable outlets, and verify requests through another channel, such as calling the person directly. Some families agree on a code word to confirm it’s really them in an emergency.
Technology companies are developing watermarks and labels to show when content has been generated by AI, but these aren’t yet used consistently.
