You can enable subtitles (captions) in the video player
It may sound like him, but this phone message, sent to voters in January, is not from Joe Biden.
Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again.
Rather, it's an AI-generated fake, encouraging recipients not to vote in New Hampshire's Democratic presidential primary.
It's important that you save your vote for the November election.
And this foul-mouthed outburst from UK Labour leader Sir Keir Starmer, berating a party staffer last year, is also an audio deepfake.
I literally told you, didn't I, [bleep], bloody moron.
By the end of this year around 2bn people will have headed to the polls, including in the US, the EU, India, and the UK. Those voters' decisions will have a significant impact at both a national and geopolitical level, as elected governments implement policies affecting matters ranging from taxation and spending, to trade and international relations.
But today it is cheaper and easier than ever to manipulate media, and potentially elections, thanks to powerful tools such as ChatGPT. These images of Donald Trump may be relatively easy to identify as fakes, but manipulated media is becoming increasingly hard to spot.
For those looking to disrupt elections, fake audio, in particular, could prove to be a potent weapon. New audio technology means AI can clone a voice from just three seconds of recordings.
In February, Amazon, Google, Meta, Microsoft, TikTok, and OpenAI were among 20 tech companies that said they would work together to combat the creation and spread of AI-generated content designed to mislead voters. Efforts could include adding audio or video watermarks to content, making it clear where it came from and whether it had been altered.
AI-generated misinformation is seen as one of the biggest global risks of 2024, according to a World Economic Forum survey, trailing only extreme weather.
Lawmakers globally have been fighting back. In February, Brazilian authorities banned the use of deepfake technology for electoral purposes. And in March, the EU introduced provisional legislation to regulate AI models and systems to comply with specific transparency obligations and EU copyright laws.
By the end of May, 44 US states had introduced or passed bills to regulate deepfakes in elections. So far, there have been no major deepfake issues at this year's elections. But as the quantity of fake AI-generated media grows, it will likely get harder for voters to believe what their ears and eyes tell them.
