The Scottish Parliament’s live video streams and recorded broadcasts are at risk from deepfake technology, according to a new report from the Scottish Centre for Crime and Justice Research in collaboration with the University of Edinburgh.

Deceptive videos and audio clips could pose a serious threat to Scottish democracy, according to the study, with the potential for sophisticated AI technologies to misuse livestreaming from the parliament’s debating chamber and committee rooms to create so-called ‘chamberfakes’.

The report, Chamberfakes: Assessing the Threats Posed by Generative AI Technologies to Parliamentary Democracy in Scotland, was commissioned by the Scottish Parliament to evaluate the potential threat of AI to parliamentary business, and to consider how deepfake videos may affect the integrity of Parliament TV.

Despite finding many examples of good practice, the team are now calling on the Scottish Parliament to go further, urging for a formal process to respond to deepfake threats and the allocation of more staff to manage the risk of attacks.

The researchers identified three main risks, including the video livestream being hacked, the dissemination of deepfakes on social media, and the many hours of parliamentary footage available online being used as feedstock to create harmful and abusive deepfakes of MSP’s.

Dr Ben Collier, co-author of the report, said: “Scotland was one of the world’s first parliaments to experiment with live streaming. While the original aim was to produce a neutral record for the public and broadcasters on what goes on inside Holyrood, it paradoxically has introduced new security risks that could threaten public trust in democracy.

“The ramifications of ‘chamberfakes’ are potentially severe but we are optimistic about the Scottish Parliament’s ability to respond to these sorts of attacks in large part due to the skilled broadcasting team who have technical processes and human checks already in place.”

To tackle the threat, the report includes three key recommendations to bolster existing checks and procedures.

The first is for the parliament to develop a plan and reporting procedures to be used during deepfake or misinformation incidents and a named person given responsibility for the task via the risk register, as currently there are no formal processes in place. 

‘Human in the loop’ and material checks are also recommended, which would involve having cameras dump their live feeds directly to a local file, as well as simple authentication checks for those dialling in, and an in-house team monitoring feeds in comparison to live proceedings.

Lastly, the report argues that a new parliamentary communications team should be established to support MSPs who fall victim to deepfake attacks, as well as tracking how content is being circulated and used online.

In May, the new first minister John Swinney was the subject of a deepfake attack when a speech purportedly showing him lambasting SNP policies went viral on social media, reportedly being seen over 300,000 times.

Recommended reading

• YouTube Introduces Tool to Flag AI Content Ahead of Global Elections
• OpenAI Is Trying to Stop AI Meddling in Elections. Will It Be Enough?
• Meta To Start Labelling AI-generated Images

Although that incident was an obvious fake due to the content, there are major concerns that more subtle, insidious deepfakes could result in the rapid spread of misinformation.   

Earlier this year, the independent fact checking non-profit, Full Fact, warned that the UK was falling behind in tackling deepfakes, and that the government’s Online Safety Act leaves several gaps in combatting the misinformation that stems from the spread of deepfakes online.  

In a year marked by an unprecedented wave of global elections there were heightened concerns over the potential surge of political deepfakes, however, new details from Meta this month showed that despite the concerns, there was limited evidence of AI being used to target voters in this way.