OpenAI stops ‘disrespectful’ Martin Luther King Jr deepfakes

OpenAI has stopped its artificial intelligence (AI) app Sora creating deepfake videos portraying Dr Martin Luther King Jr, following a request from his estate.

The company acknowledged the video generator had created “disrespectful” content about the civil rights campaigner.

Sora has gone viral in the US due to its ability to make hyper-realistic videos, which has led to people sharing faked scenes of deceased celebrities and historical figures in bizarre and often offensive scenarios.

OpenAI said it would pause images of Dr King “as it strengthens guardrails for historical figures” – but it continues to allow people to make clips of other high profile individuals.

That approach has proved controversial, as videos featuring figures such as President John F. Kennedy, Queen Elizabeth II and Professor Stephen Hawking have been shared widely online.

It led Zelda Williams, the daughter of Robin Williams, to ask people to stop sending her AI-generated videos of her father, the celebrated US actor and comic who died in 2014.

Bernice A. King, the daughter of the late Dr King, later made a similar public plea, writing online: “I concur concerning my father. Please stop.”

Among the AI-generated videos depicting the civil rights campaigner were some editing his infamous “I Have a Dream” speech in various ways, with the Washington Post reporting one clip showed him making racist noises.

Meanwhile others shared on the Sora app and across social media showed figures resembling Dr King and fellow civil rights campaigner Malcolm X fighting one another.

AI ethicist and author Olivia Gambelin told the BBC OpenAI limiting further use of Dr King’s image was “a good step forward”.

But she said the company should have put measures in place from the start – rather than take a “trial and error by firehose” approach to rolling out such technology.

She said the ability to create deepfakes of deceased historical figures did not just speak to a “lack of respect” towards them, but also posed further dangers for people’s understanding of real and fake content.

“It plays too closely with trying to rewrite aspects of history,” she said.

The rise of deepfakes – videos which have been altered using AI tools or other tech to show someone speaking or behaving in a way they did not – have sparked concerns they could be used to spread disinformation, discrimination or abuse.

OpenAI said on Friday while it believed there were “strong free speech interests in depicting historical figures”, they and their families should have control over their likenesses.

“Authorised representatives or estate owners can request that their likeness not be used in Sora cameos,” it said.

So-called “cameos” on the platform allow living people to consent to having their face or resemblance used in further AI videos on Sora.

OpenAI told the BBC in a statement in early October it had built “multiple layers of protection to prevent misuse”.

And it said it was in “direct dialogue with public figures and content owners to gather feedback on what controls they want” with a view to reflecting this in subsequent changes.