The U.S. and EU Brace for AI-Powered Election Disinformation

(Getty Images)

Nascent U.S. and EU regulations and nonbinding private sector efforts on artificial intelligence (AI) tools will likely fail to meaningfully control the influx of AI content on social media platforms during the 2024 elections in both territories, opening the door to the proliferation of malign synthetic content. In the past several weeks, governments and technology companies have intensified their focus on the potentially negative impact of generative AI tools on election processes. On March 14, the European Commission requested information from eight major social media platforms on their efforts to mitigate risks to electoral processes linked to generative AI, including so-called ''hallucinations'' (when chatbots output false responses) and ''deepfakes'' (malign synthetic audio or visual content meant to deceive audiences). A month earlier, at the Munich Security Conference in February 2024, 20 leading technology companies, including Amazon, Anthropic, Meta, Microsoft, OpenAI, Snap, Stability AI, TikTok and X, signed a series of commitments to work together to detect and counter harmful AI content, particularly in regards to elections, pledging to deploy technology countering harmful AI-generated content meant to deceive voters and mitigate the presence of deceptive AI content on social media platforms. These companies also pledged to create tools like watermarks and detection techniques to spot and label synthetic content, as well as stress-testing AI models to evaluate how they could be leveraged to disrupt elections.

The European Commission sent requests for information under the European Union's Digital Services Act (DSA) to Bing, Google Search, Facebook, Instagram, Snapchat, TikTok, YouTube, and X. The companies will have until April 5 to disclose how generative AI content is created and spread on their platforms and services, and how they identify and mitigate deepfakes and other malign synthetic content.

Although the European Union has some legislation in place to regulate online content and AI tools, these laws are either nascent or have not yet been enforced, while in the United States, there is no federal AI framework or laws pertaining to synthetic content use cases. In August 2023, the European Union's DSA came into effect, introducing a comprehensive set of regulations for large platforms like social media companies, online marketplaces and search engines. The DSA requires these companies to identify, analyze and assess systemic risks that are linked to their services, including in relation to electoral processes. The law also requires companies to remove illegal content like misinformation and disinformation, though the DSA does not extend to AI applications. The European Union's much more sweeping AI framework, the Artificial Intelligence Act, was adopted by the European Parliament on March 13. However, the AI law will not enter into force until May or June, and the requirements for companies will not be fully applicable until mid-2026. Meanwhile, in the United States, there is no federal legislation governing the use of AI or laws banning deepfake technology, and the prospects of one being approved are very low in Washington's polarized political climate. While an array of U.S. states have introduced bills pertaining to AI and synthetic content regulation, particularly during election periods, many of these efforts are nascent and varied, undermining their utility. Some U.S. regulatory agencies have taken steps to put restrictions on the use of AI tools. Most recently, the Federal Communications Commission banned the use of AI-generated voices in robocalls in February. However, these efforts have also had a limited scope and application; they do not, for example, apply to audio synthetic content if such content circulates on social media platforms or in campaign advertisements.

Under the European Union's DSA, illegal online content is banned and punishable with fines as high as 6% of a company's global turnover, and companies could be banned from operating in the bloc if they repeatedly fail to comply. On March 14, the European Union announced a formal investigation into the Chinese e-commerce site AliExpress over potential violations of the DSA's rules.
In 2024, U.S. lawmakers in 32 states have so far introduced 52 bills to regulate deepfakes in elections, according to the U.S. nonprofit advocacy group Public Citizen.

The incipient AI regulations in the European Union and the United States, combined with nonbinding private sector pledges, will likely be insufficient in compelling social media companies to curb the surge of generative AI content being shared on their platforms in the short-to-medium term. The European Union is attempting to show it has teeth behind the DSA, as evidenced by its announced probes into Meta and TikTok in October 2023 following the onset of the Israel-Hamas war and a subsequent surge of disinformation and other ''illegal content'' on the two companies' social media platforms. However, the EU regulation is still less than a year old and its full enforcement capabilities remain largely untested. The much more applicable EU AI Act will also fail to oversee and mitigate AI use cases during the June elections for the European Parliament because of the law's deferred enforcement. Meanwhile, the absence of any federal legislation governing AI uses in the United States will fall short of any meaningful oversight and regulation over the private sector's AI mitigation efforts, and how U.S. presidential candidates wield AI tools ahead of the November election. In the private sector, even with many organizations signing on to international AI pledges, the ambiguity in the wording of these accords and the lack of binding requirements will also likely undermine tangible efforts to counter disinformation by companies. And while the Western companies driving the AI sector, including Google, Meta and OpenAI, have all pledged to introduce safeguards on generative AI creators, many of these proposed transparency tools have not yet been rolled out and may still be evadeable via workarounds. These companies have other election policies in place, such as Meta, which claims its content moderators remove misinformation about ''the dates, locations, times, and methods for voting, voter registration, or census participation'' and false posts purportedly intended to interfere with someone's civic participation. However, such policies have so far failed to curb the spread of disinformation, which has run rampant during several recent international crises, including the Israel-Hamas war.

As a result of these insufficient safeguards, synthetic content will likely proliferate on social media platforms ahead of the EU and U.S. elections, amplifying the appeal of disinformation and inflaming social grievances in a number of countries. More convincing deepfake content will likely compound safety and logistical risks already present in some Western countries (for example, European countries grappling with ongoing farmers' protests). Candidates and political actors will likely leverage AI tools to strengthen their image or deface their opponents. Other threat actors, including financially motivated cybercriminals, may also wield synthetic audio content using election-themed lures to dupe citizens into disclosing sensitive financial information or login credentials. In the long term, ongoing public and private sector initiatives are poised to bolster platforms' resilience to malign synthetic content, which will likely pose a less severe threat in future election cycles.

In February, Meta announced it would start labeling AI-generated images that users post to Facebook, Instagram and Threads ''in the coming months.'' Google made a similar announcement the same month, stating that it was looking to create an AI watermark tool that would provide information about when, where and how the media was edited.