[SCS0028]
Written evidence submitted by the Center for Countering Digital Hate
The Center for Countering Digital Hate (CCDH) is grateful for the opportunity to submit evidence to this important inquiry.
CCDH is a non-profit research and advocacy organisation with offices in London and Washington DC. CCDH conducts independent studies, campaigns for social media reform, and advises lawmakers and regulators on the basis of our research. The following submission is drawn from CCDH’s research into online harms. Particularly relevant to this inquiry is research we published in August 2024 on the abuse received by female politicians.
Answers to the inquiry’s specific questions are below.
The lack of functional safeguards for political candidates on social media channels is a significant risk factor influencing the level of threat they face.
Social media has revolutionised politics. Online spaces are now the primary places where societal norms and values are negotiated and normalized, and where we learn about and discuss current events, social issues, and politics. Through social media, political candidates have a direct line of communication to voters and constituents. But these campaigning and communication advantages come with significant risks to safety.
Social media platforms say they take the safety of politicians and the integrity of elections seriously, but CCDH research has shown how platforms fail to step up to protect candidates and public officials and decline to uphold their own rules on abusive content. These factors are compounding: abusive, violent rhetoric on social media increases the hostility of the political environment. This in turn fosters an environment where threats are normalised and flow ever more freely. Threatening comments are publicly visible, being seen by supporters and opponents alike and having knock-on effects on other social media users. Individuals who might otherwise be interested in public office witness the torrent of abuse candidates face and consider whether it is really worth it.
To quantify the nature of this abuse, and the threat it portends to political candidates, in 2024 CCDH collected 560,000 comments on Instagram posts from five leading female politicians. Using a machine learning tool, we assessed that 1 in 25 replies to these politicians’ posts were toxic, including heinous comments containing sexist abuse, racist abuse and death threats. CCDH reported all abusive comments to the platform but found that Instagram failed to act on 92% of the reports.
And it's not just Instagram. A 2022 BBC investigation found that more than 3,000 offensive tweets are sent to British MPs every day. We find the same lack of safeguards for other public figures, partnering with influencers and celebrities to reveal how platforms fail to act on 9 in 10 reports of misogyny in direct messages. The safeguards are few and the accountability non-existent; but for politicians and candidates, these online threats can have very real offline consequences.
A lack of accountability is a significant driver of the abusive online behaviour that endangers political candidates. Campaigning offline, be it on the doorstep, at a community centre or political rally, we would not tolerate violent, racist, or misogynistic slurs being hurled at candidates seeking to serve their community in public office. An abuser would be thrown out of the venue, fast. Yet on social media, an abuser can barrage a candidate with rape and death threats and then continue to use the platform with impunity. In CCDH’s research on the abuse faced by women candidates, we found that a fifth of toxic comments came from “repeat offenders”, users who had sent abuse to the candidate more than once.
Also driving this behaviour is the social media business model which relies on user engagement to increase advertising revenue. This emphasis on ever-increasing engagement has created a malignant dynamic where incendiary, contentious commentary is rewarded by a platform’s algorithm, earning both platform and user financial rewards for their toxic posts. This distorts political discourse, with financially motivated users directing abusive content at candidates and public figures for economic reward.
False information -- intentional (disinformation) or unintentional (misinformation) -- have serious risks to democratic processes and the safety of candidates. CCDH does not believe there are sufficient measures in place to identify, tackle or deter such material.
False and hateful information online has been a component radicalising individuals who later turn to political violence. Social media platforms should take that fact seriously and take steps to address the circulation of false information with the potential to radicalise.
Sadly, major platforms seem to be heading in the opposite direction. On 7 January 2025, Meta announced policy changes to the way it handles false information, removing independent fact-checks and stopping proactive enforcement on some kinds of harmful content. The seismic change this bodes for users of Meta platforms appears, to CCDH, to greatly raise the risks discussed here.
The riots which broke out following the tragic Southport stabbings are another example of the dangerous risks from false information on social media. CCDH research showed how false information concerning the identity of the attacker spread rapidly across social media, promoted to users by platform algorithms and recommender features.
All this occurred shortly after the UK General Election on 4 July 2024. The impact of this online misinformation, and the violence it spawned, has been profound. Democracies rely on trust and citizens’ shared sense of civic duty. Following the riots, these necessary ingredients for a healthy democracy are battered.
CCDH has been researching the risks generative AI tools pose to democratic processes. We have shown how voice cloning tools and image generators can be used to create misleading pictures and audio recordings featuring political candidates.
We found that popular AI audio cloning tools could be easily manipulated to create dangerous false audio clips in the voices of high-profile politicians, including UK politicians Rishi Sunak and Keir Starmer. Voice cloning has been used to influence elections globally, often in ways that purposefully attempt to discourage voters. Incidents include:
Candidates have also been falsely endorsed through voice cloning. A candidate in India was falsely endorsed by the voice clone of a deceased former party leader, and a candidate in Taiwan was falsely endorsed by the AI voice of former candidate and billionaire Terry Gou. These examples show how AI poses to elections via the manipulation of voter opinion and casting doubt on the integrity of electoral processes.
In summary, generative artificial intelligence tools and large language models have vast potential to play a considerable role in the creation and spread of misinformation, disinformation and harmful content.
Social media companies must uphold their own rules on abusive and hateful content, including that directed at political candidates. Ofcom must strongly enforce the requirements of the Online Safety Act, which include the duty to uphold terms and conditions and empower users with recourse pathways. Reforms to digital advertising markets must continue to remove the financial rewards for content creators directing hateful comments at political candidates. Social media platforms should also undertake risk assessment of their impact on democratic processes, elections and candidate safety, and publish these for independent analysis.
CCDH thanks the Speaker’s Conference for undertaking this important inquiry.
7 February 2025
| 3 |
|