AI Deepfakes in Maryland Elections: Officials Report Low Impact

As artificial intelligence tools proliferate, concerns about deepfake manipulation in elections have intensified. Maryland election officials recently reported a notable absence of AI-generated deepfakes in their jurisdiction, raising questions about the scale of the threat and the effectiveness of existing safeguards. This investigation examines the data, the broader narrative around deepfakes, and what it means for voter trust and election integrity.

In an era where misinformation can sway electoral outcomes, the claim that Maryland election officials have observed few AI deepfakes in recent elections warrants scrutiny. According to a report by GovTech, officials in Maryland have identified only a handful of suspected AI-generated audio or video clips targeting voters or candidates. This stands in contrast to the widespread alarm about deepfakes in elections, which has been amplified by media outlets, cybersecurity firms, and even government agencies. The discrepancy between perceived risk and reported incidents demands a closer look at the data, the methods used to detect deepfakes, and the potential implications for election security.

Context and Background on AI in Elections

The integration of artificial intelligence into electoral processes has introduced both efficiencies and vulnerabilities. AI tools are increasingly used for voter registration verification, fraud detection, and even candidate outreach. However, the same technologies that streamline operations can also be weaponized to create convincing but fabricated content. Deepfakes—hyperrealistic audio or video manipulations—represent one of the most insidious applications of AI in this context, as they can undermine public trust by spreading false narratives about candidates, policies, or electoral processes.

Historically, election interference has relied on traditional disinformation tactics, such as coordinated social media campaigns or leaked documents. The advent of deepfakes, however, introduces a new dimension: the ability to fabricate entire events or alter the words of public figures with near-perfect authenticity. This capability has been demonstrated in high-profile cases, including the spread of AI-generated audio clips of political leaders making false statements. While these incidents have often occurred in other jurisdictions, the question remains whether Maryland—like other states—has been spared due to proactive measures, limited targeting, or simply the absence of large-scale campaigns.

Why Deepfakes Pose a Unique Threat

• Perceived Authenticity: Deepfakes leverage advanced machine learning to replicate voices, facial expressions, and even body language, making them difficult to distinguish from genuine content without technical analysis.
• Scalability: Unlike traditional disinformation, which requires manual production, deepfakes can be generated en masse with minimal resources, enabling rapid dissemination across digital platforms.
• Psychological Impact: The human brain is wired to trust visual and auditory cues, particularly when they appear to come from trusted sources. A deepfake of a candidate making a controversial statement can erode voter confidence in the electoral process itself.

The potential for deepfakes to disrupt elections has been a recurring theme in cybersecurity and electoral integrity discussions. For instance, the Reuters Institute for the Study of Journalism has highlighted how deepfakes could be used to manipulate public opinion by creating plausible but false narratives about candidates’ past actions or policy positions. While Maryland officials report minimal deepfake activity, the broader context suggests that this could be a matter of degree rather than absence of risk.

The Emerging Narrative Around Deepfakes

The discourse around deepfakes in elections has been shaped by a mix of real-world incidents, speculative warnings, and industry-driven alarmism. Media coverage often frames deepfakes as an imminent existential threat to democracy, citing examples from other countries where AI-generated content has been used to influence voters. However, the narrative is not always grounded in empirical data, particularly when it comes to localized election cycles like Maryland’s.

Cybersecurity firms and think tanks have played a significant role in amplifying concerns about deepfakes. For example, Bloomberg has reported on how state-sponsored actors in countries like Russia and China have experimented with deepfake technology to sow discord in foreign elections. While these cases underscore the potential for deepfakes to be used as a tool of foreign interference, they do not necessarily reflect the tactics or priorities of domestic actors in jurisdictions like Maryland.

Media Framing and the “Deepfake Hype”

The media’s portrayal of deepfakes often emphasizes their potential rather than their actual deployment. Headlines frequently describe deepfakes as “the next frontier of election interference,” even when concrete examples are scarce. This framing can create a self-fulfilling prophecy, where the anticipation of deepfakes leads to heightened scrutiny and, paradoxically, more opportunities for misinformation to spread under the guise of “preventive” measures.

It is worth noting that the narrative around deepfakes is not uniform. Some analysts argue that the technology is still nascent and that its effectiveness in large-scale disinformation campaigns has been overstated. Others point to the fact that deepfakes require significant resources to produce and distribute effectively, which may limit their use in smaller or less resourced campaigns. Maryland’s reported lack of deepfake activity could thus reflect both the technical barriers to large-scale deployment and the effectiveness of local safeguards.

What GovTech Data Shows in Maryland

The report from GovTech provides specific details about the nature and extent of AI deepfake activity in Maryland elections. According to the report, officials have identified only a few instances where AI-generated content appeared to target voters or candidates. These cases were flagged through a combination of automated tools and human review, with most deepfakes being detected in the final stages of dissemination rather than during production.

The report does not specify the exact number of deepfakes identified, but it emphasizes that none of the detected content had a measurable impact on voter behavior or election outcomes. This aligns with broader trends in election security, where the majority of disinformation campaigns—whether AI-generated or not—fail to sway large numbers of voters. However, the absence of deepfakes does not necessarily indicate a lack of threat; it may instead reflect the effectiveness of monitoring systems or the absence of a coordinated campaign targeting Maryland.

Detection Methods and Limitations

Maryland officials have attributed their ability to detect deepfakes to a multi-layered approach that includes:

• Automated Scanning: Use of AI-driven tools to flag suspicious content on social media and other platforms.
• Human Review: Dedicated teams of analysts who cross-reference flagged content with known sources and verify its authenticity.
• Partnerships with Tech Companies: Collaboration with platforms like Facebook, Twitter (now X), and YouTube to remove or label deepfake content.

However, the report does not provide details on the false positives or negatives associated with these methods. For instance, automated tools may occasionally misclassify legitimate content as deepfakes, leading to unnecessary scrutiny or removal. Conversely, some deepfakes may evade detection due to improvements in AI generation techniques or the use of less sophisticated but still effective manipulation tools.

Assessing the Actual Impact on Voters

The question of whether deepfakes have had any impact on Maryland voters cannot be answered definitively based solely on the number of detected instances. Impact is a multifaceted concept that includes not only measurable changes in voter behavior but also the erosion of trust in the electoral process. Even if deepfakes have not swayed large numbers of voters, their mere presence could contribute to a broader climate of distrust, particularly if voters perceive that election officials are unable to protect them from misinformation.

Research from the Pew Research Center suggests that voters are increasingly skeptical of political content online, regardless of whether it is manipulated or not. This skepticism may be exacerbated by the presence of deepfakes, even if they are rare. For example, a single high-profile deepfake—even if debunked—could reinforce the perception that the election is being manipulated, thereby undermining confidence in the process as a whole.

A Comparison of Claims vs. Evidence

The table above highlights the disconnect between the alarmist claims often heard in media discussions and the empirical data from Maryland. While the threat of deepfakes is real, the actual deployment in Maryland appears limited, at least in terms of detectable instances. This does not mean the threat is nonexistent, but it does suggest that the risk may be overstated in some contexts.

Institutional Monitoring and Preparedness

Maryland’s approach to monitoring and mitigating deepfake threats is indicative of a broader trend in election security: a reliance on a combination of technology, human expertise, and interagency collaboration. The state’s election officials have implemented several measures to detect and respond to deepfakes, including:

• Establishment of a Cybersecurity Task Force dedicated to tracking disinformation and deepfake activity.
• Integration of AI-driven content analysis tools to preemptively identify suspicious material.
• Development of public awareness campaigns to educate voters on recognizing deepfakes.

However, the effectiveness of these measures is difficult to quantify. While automated tools can flag potential deepfakes, they are not infallible. Human review is essential to avoid false positives, but it also introduces delays and potential biases. Additionally, the task force’s ability to respond to deepfakes depends on real-time collaboration with social media platforms, which may not always be forthcoming or consistent.

Challenges in Monitoring Deepfakes

One of the primary challenges in detecting deepfakes is the rapid evolution of AI generation tools. As deepfake technology improves, so too do the methods used to detect it. Maryland officials have not disclosed whether they have faced difficulties in keeping up with these advancements. Similarly, the lack of standardized definitions for what constitutes a deepfake can lead to inconsistencies in reporting and detection.

Another challenge is the siloed nature of data collection. Deepfakes may originate from private platforms, foreign servers, or even dark web forums, making it difficult for election officials to track their spread. Without comprehensive data-sharing agreements with tech companies and international partners, gaps in monitoring are inevitable.

Fact-Checking the Digital Threat Landscape

Fact-checking deepfakes presents unique challenges due to the nature of the content itself. Unlike traditional misinformation, which can often be traced to a specific source, deepfakes may appear to originate from legitimate accounts or organizations. This makes verification more difficult and time-consuming, particularly when fact-checkers are already stretched thin.

Organizations like PolitiFact and Snopes have developed methodologies for evaluating deepfakes, including:

• Metadata Analysis: Examining technical artifacts in audio or video files to determine their origin and potential manipulation.
• Reverse Image Search: Using tools like Google Reverse Image Search to identify whether a video or image has been previously published or altered.
• Expert Consultation: Engaging with computer vision and audio forensics specialists to assess the authenticity of suspicious content.

In Maryland, the burden of fact-checking deepfakes falls primarily on election officials and, to a lesser extent, on independent fact-checkers. The lack of a centralized, well-funded fact-checking infrastructure for deepfakes means that verification efforts may be reactive rather than proactive. This reactive approach can leave voters exposed to misinformation until after it has been debunked, potentially allowing it to spread undetected.

The Role of Platforms in Mitigating Deepfakes

Social media platforms play a critical role in the detection and removal of deepfakes. Companies like Facebook, Twitter, and YouTube have implemented policies to label or remove AI-generated content that violates their terms of service. However, enforcement is inconsistent, and some deepfakes may evade detection due to their subtle nature or the use of alternative platforms.

The CNN has reported that platforms often struggle to balance free speech concerns with the need to prevent harm from deepfakes. For example, a deepfake that appears to show a candidate making a false statement may be removed if it is deemed to violate platform rules, but the decision process is not always transparent. This lack of clarity can undermine public trust in the platforms themselves, further complicating the fight against misinformation.

Preventative Measures Taken by Officials

In response to the potential threat of deepfakes, Maryland election officials have taken several preventative measures to safeguard the integrity of the electoral process. These measures are designed to both deter potential attackers and mitigate the impact of any deepfakes that do emerge. Key initiatives include:

Enhanced Voter Education: Campaigns to inform voters about the risks of deepfakes and how to verify the authenticity of political content. This includes partnerships with local media outlets and educational institutions to disseminate reliable information.

Secure Communication Channels: Implementation of encrypted or verified communication methods for candidates and officials to reduce the risk of impersonation via deepfake audio or video.

Collaboration with Tech Companies: Working with platforms to develop and deploy tools that can detect and flag deepfakes in real time. This includes sharing data on emerging deepfake techniques to help platforms stay ahead of new threats.

However, preventative measures are not without limitations. For instance, voter education efforts may be ineffective if voters lack the technical literacy to recognize deepfakes. Similarly, secure communication channels can be bypassed if attackers target other points of vulnerability, such as social media accounts or public speeches.

Legal and Policy Responses

At the state and federal levels, there have been discussions about enacting laws specifically targeting deepfakes in elections. For example, some states have passed legislation making it illegal to create or distribute deepfakes with the intent to influence voters. Maryland has not yet introduced such legislation, but officials have expressed support for federal efforts to address the issue.

The effectiveness of legal responses depends on several factors, including:

• Enforcement Capabilities: Law enforcement agencies must have the resources and expertise to investigate and prosecute deepfake-related crimes.
• International Cooperation: Deepfakes can originate from anywhere in the world, making cross-border cooperation essential to track and dismantle networks responsible for their creation and distribution.
• Public Awareness: Laws alone are not sufficient; voters must be informed about their rights and the legal consequences of spreading deepfakes.

Red Flags Checklist: Identifying Potential Deepfakes

While election officials play a crucial role in detecting deepfakes, voters can also take steps to identify suspicious content. Below is a checklist of red flags to watch for when evaluating political audio or video:

• Unnatural Blinking or Eye Movement: Deepfakes often struggle to replicate subtle facial movements, such as blinking or eye tracking, which can appear robotic or inconsistent.
• Lip Sync Issues: The audio and visual elements may not align perfectly, particularly if the deepfake was generated using separate audio and video tracks.
• Unrealistic Shadows or Lighting: Deepfakes may produce shadows or lighting effects that do not match the real-world environment depicted in the video.
• Background Noise or Artifacts: Some deepfakes introduce digital artifacts, such as graininess or compression errors, which can be a sign of manipulation.
• Sudden Changes in Tone or Expression: If a person’s voice or facial expressions shift abruptly without context, it could indicate editing or AI manipulation.
• Lack of Context: Deepfakes are often shared without clear sourcing or context, particularly if they are designed to spread quickly before verification is possible.
• Platform Warnings or Labels: Many social media platforms now label AI-generated content. If a video or audio clip is marked as “potentially altered” or “synthetic media,” it is a strong indicator that it may be a deepfake.

Voters should also cross-reference suspicious content with trusted news sources and fact-checking organizations before sharing or acting on it. When in doubt, it is safer to assume that the content may be manipulated and seek additional verification.

Frequently Asked Questions

Are deepfakes a significant threat to Maryland elections?

Based on reports from Maryland election officials, deepfakes have not yet posed a significant threat to the state’s elections. However, the absence of detected deepfakes does not mean the threat is nonexistent. Officials emphasize that monitoring and prevention efforts are ongoing, and the risk could change with advancements in AI technology or shifts in adversarial tactics.

How do election officials detect deepfakes?

Maryland officials use a combination of automated tools and human review to detect deepfakes. Automated systems scan social media and other platforms for suspicious content, while dedicated teams of analysts verify flagged material using technical analysis and cross-referencing with known sources. Collaboration with tech companies also plays a key role in identifying and removing deepfake content.

Can voters recognize deepfakes on their own?

While some deepfakes may exhibit telltale signs of manipulation—such as unnatural facial movements or lip sync issues—most require technical expertise to detect. Voters can reduce their risk of encountering deepfakes by staying informed about common red flags, verifying content through trusted sources, and being skeptical of highly emotional or sensational political media.

What should voters do if they encounter a suspected deepfake?

If a voter encounters content they believe may be a deepfake, they should not share it without verification. Instead, they can report the content to the platform where it was found, consult fact-checking organizations, and consult official election channels for guidance. In Maryland, voters can also reach out to local election officials for assistance in evaluating suspicious content.

Are there laws against using deepfakes in elections?

Several states, including some neighboring Maryland, have enacted laws specifically targeting the use of deepfakes to deceive voters or influence elections. At the federal level, there have been proposals to address deepfakes, but no comprehensive legislation has been passed as of yet. Maryland officials support federal efforts to close legal gaps and enhance penalties for those who misuse deepfake technology in electoral contexts.

How can tech companies help prevent deepfakes in elections?

Tech companies can play a critical role in preventing deepfakes by implementing robust detection tools, labeling AI-generated content, and collaborating with election officials to share intelligence on emerging threats. Platforms can also invest in research and development to improve their ability to distinguish between real and synthetic media. Additionally, companies can educate users about the risks of deepfakes and provide clear guidelines for reporting suspicious content.

Sources & References

• GovTech — Maryland Election Officials Reporting Few AI Deepfakes