Defending the Homeland Against AI

Driven Information Warfare

By Capt. Nolan R. Hedglin

Article published on: September 1, 2025 in the Gray Space Fall 2025 Edition

Read Time: < 10 mins

Introduction

Artificial intelligence (AI), which is broadly defined as the ability to stimulate human cognition through machine action, can empower individuals and organizations to create and disseminate digital advertising at an unprecedented scale. Through advanced techniques in profiling individuals from their online behavior, organizations can wage micro-targeted influence operations—be it to influence a user’s purchasing decisions or political activity—automatically with minimal resources and across borders. The United States (U.S.) has contended with this reality in every election since 2018 after The New York Times reported that Cambridge Analytica misused Facebook marketing data to conduct targeted political advertising in the 2016 U.S. election (Hakim & Rosenberg, 2018). Historically, AI’s role in micro-targeting has been limited to individual profiling. Within the past year, however, platforms such as Meta, Amazon, and TikTok are experimenting with allowing advertisers to use generative AI when creating content (Roth, 2023). By arming individual advertisers with tools to automatically generate ads, these platforms fully connect the stack required to identify a new target and influence their behavior without human input.

Malign influence actors (MIAs) ride on the backbone of commercially available data and profiling tactics, and the introduction of generative AI into commercial ad-tech stacks should be perceived as a fundamental shift in how malign influence activity may occur in the future. In this paper, I argue the U.S. should re-focus efforts away from adopting a “defend-forward” framework for combatting the foreign influence of U.S. citizens. Taking active measures to counter malign influence activity may be less effective than in the past, as the final step of influencing an audience - content creation and delivery - no longer relies on human input. Actively defending automated influence campaigns would require an automated system of our own. Furthermore, the U.S. is not effectively equipped to combat foreign influence domestically in an automated manner that would be compliant with federal privacy laws. Instead, I propose that the U.S. should partner with the European Union (EU) and conduct a case study on differing governance models regarding the protection of consumer privacy. This would aid U.S. legislators in enacting a federal consumer privacy law that curbs unfettered data collection against U.S. citizens. By providing stronger protections against violations of consumer privacy, the U.S. would severely inhibit a malign influence actor’s ability to profile U.S. citizens and serve AI-developed content.

Generative AI Exponentially Increases the Rate of Malign Influence Activity

Two main applications arise in the assessment of the use of AI as a tool in malign influence activity: target identification and content creation (Hunter et al., 2024). Ultimately, AI rapidly accelerates the notification-to-strike decision cycle for targeted influence operations.

With respect to target identification, MIAs employ AI in the exact same manner as advertisers that conduct surveillance capitalism (Hunter et al., 2024). As an example, a website fingerprints an individual’s activities on their platform. A data broker then aggregates and connects user behavior across platforms through their advertising identity (Ramirez et al., 2014). Advertising firms analyze a user’s digital pattern of life, categorizing them into over 70 buckets via clustering (e.g. person X is a “tech enthusiast”) and predicting their response to new advertisements through supervised learning techniques such as extreme gradient boosting and recurrent neural networks (RNNs) (Chen & Guestrin, 2016; Ebadi Jokandan et al., 2022). Ostensibly, MIAs should not find much use from an advertiser’s robust profiling of consumer behavior. However, reports indicate that consumer behavior reveals political behavior. Thus, consumer behavior is an inroad for MIAs to target susceptible audiences (Jung & Mittal, 2020).

With respect to content creation, MIAs use AI in many forms. For example, generative adversarial networks (GANs) create deepfake videos through tools such as DeepFaceLab (Liu et al., 2023). Variational autoencoders (VAEs), like DALL-E, provide near-instantaneous text-to-image generation (Ramesh et al., 2022). Lastly, large language models (LLMs) such as Open AI’s GPT-4 can mimic human speech across multiple languages with minimal operator input (Naveed et al., 2024). As a kit of content creation tools, AI can enable MIAs to create and deliver content at machine speeds.

Although AI may hallucinate while creating content or misidentify vulnerable targets, MIAs do not view this as a major concern. Through the command and control of a bot, AI can deliver strikes at a rate that far outpaces any manual action. In an effective information warfare campaign, MIAs employ perpetual information-barraging during competition to gradually change the attitudes and perceptions of a target audience. These actions then set the conditions for precision strikes to affect an individual’s decision cycle at the optimal moment, such as who they should vote for in an upcoming election (Hunter et al., 2024).

To exacerbate matters further, introducing generative AI into the ad-tech stack means every user may encounter bespoke advertisement. This presents a unique challenge to actively defend against influence campaigns. A core principle in understanding the intentions and tactics of MIAs is by comparing which target audiences are receiving the same messages. Developing a broad situational understanding of who a MIA chooses to target allows the U.S. to appropriately allocate resources to defend acute threats. However, if users are consistently served unique content, the tools we typically use to cross-reference influence activities may become ineffective.

The U.S. is Ill-Equipped to Automatically Combat Foreign Influence Domestically

The U.S. has developed strong guardrails around data collection on its citizens as compared to Russia and China, which puts the DoD at a severe disadvantage in controlling the information environment domestically. Russia and China leverage their authoritarian structure to collect vast amounts of data on their citizens without consent (Hunter et al., 2024). These regimes then use surveillance capitalism techniques to micro- target citizen behaviors and execute AI-driven censorship campaigns domestically, akin to a “defend-forward” construct within the information environment (Dawson, 2021). To employ similar tactics domestically, the U.S. would have to become a master of two functions: citizen profiling and content censorship.

As illustrated by the techniques employed in surveillance capitalism, effective citizen profiling requires persistent data collection on individual users without continually receiving their explicit consent. However, robust U.S. digital privacy laws protect U.S. citizens from government surveillance, except when evidence ultimately leads to a law enforcement action (Records Maintained on Individuals, 1988; Wire and Electronic Communications Interception, 1986). Nonetheless, the U.S. should not loosen these restrictions because such laws set the foundation of our democracy and establish trust between citizens and the government.

Executing content deletion in a “defend-forward” construct, meanwhile, violates the First Amendment because it involves government overreach into bot deletion and censorship. Additionally, a preponderance of malign influence activity occurs on social media platforms—such as Facebook, Twitter, and Reddit. Except in the case of criminal activity, federal statute treats such platforms as U.S. persons and protects them from government censorship of their platform via the First Amendment (“United States Person” Defined, 1992).Currently, the U.S. government combats malign influence activity as a reactionary measure by countering instances of disinformation and promoting government transparency. Unfortunately, experts report that presenting the truth as a reactionary measure does not sufficiently combat a barrage of foreign influence because narratives establish a foothold through emotion (Heslen, 2020). Rather than reacting to malign influence activity with truth, the U.S. ought to focus on preventing citizens from falling victim to malign influence when it is initially encountered.

A Call for Robust Consumer Privacy Protection

Protecting U.S. citizens from malign influence begins with protection from targeted profiling. Given surveillance capitalism drives targeted influence operations, the U.S. largely leaves its citizens to defend themselves against violations of consumer privacy. Although members of Congress have proposed several privacy bills, the U.S. has not enacted a single comprehensive federal privacy law to protect citizens against violations of consumer privacy, nor has Congress passed a law to regulate the behavior of data brokers. This presents a problem because data brokers and online platforms expand the threat landscape by feeding profiling algorithms with real-time information about user perceptions and attitudes.

Current federal policy (or lack thereof) has resulted in U.S. citizens taking privacy into their own hands through the FTC’s notice-and-consent mandate. Before fingerprinting occurs, a platform must notify a user of their data collection practices and request consent. Ultimately, notice- and-consent has proven ineffective because individual users do not value privacy as much as convenience (Norman-Webler, 2024). For example, in 2025 Congress banned TikTok from the U.S. market after conducting a thorough review of their consumer privacy policies. Mere hours after the ban took effect, an alternative Chinese app called RedNote with similarly problematic privacy policies became the top-downloaded app (Cheung et al., 2025).

I recommend that the U.S. initiate a partnership with the European Union to study how consumer privacy affects malign influence activity. In 2016, the EU armed citizens of its member countries with the ability to file a civil lawsuit against companies, known as a private right of action, for violations of their privacy through the General Data Protection Regulation (Art. 82 GDPR). Under the GDPR, any EU citizen can bring forth a lawsuit if they suspect that a company has: collected personal data without their consent, demonstrated negligence in protecting user data, or failed to comply with data deletion requests. A private right of action deters the practice of third-party data scraping and dramatically restricts the ability of platforms to fingerprint and sell user information without fear of economic repercussion. Presently, only the California Consumer Privacy Act (2018) contains a similar statute, though its private right of action limits lawsuits filed by California citizens to data breaches due to poor data security practices (Buchanan & Cruz, 2019).

At a high level, the U.S. and EU practice fundamentally different models of governance regarding the protection of consumer privacy, with the latter leaning heavily on top-down legislation to restrict individual data collection. And although EU legislators did not design the GDPR to curb malign influence activity, its effects may still be present in a potential case study since surveillance capitalism and malign influence activity involve the same practices. The U.S. and EU could compare data on the effectiveness of malign influence activity over the past decade within their respective countries to determine if either model of governance is more effective at preventing individual profiling by MIAs. This partnership provides an alternate path to adopting the authoritarian tactics used by Russia and China in actively combatting foreign influence.

Conclusion

Advancements in AI allow organizations to conduct influence operations at an unprecedented scale with precision down to the individual level. In the first phase of influence operations, data collection and profiling, MIAs adopt a surveillance capitalism framework to identify individual perception and attitudes. In the second phase, MIAs use generative AI to create and deliver content nearly instantaneously. To combat malign influence activity in a “defend-forward” construct, the U.S. would need to adopt citizen profiling techniques and censorship policies such as Russia and China. However, such policies do not adhere to the Fourth and First Amendment, respectively. Rather than enable government censorship, the U.S. should seek to strengthen consumer privacy protections through a comprehensive federal privacy law to degrade the target acquisition cycle for malign influence activity.

References