AI Propaganda Forces Oversight, Fuels Official Evasion
EU AI Act and Provenance Tracking Emerge
The EU AI Act, which entered into force on August 1, 2024, and became applicable on August 2, 2026, mandates transparency and disclosure for generative AI providers. Guillén Yparrea and Iván Miguel García López, along with Alexander Romanishyn, Olena Malytska, and Vitaliy Goncharuk, documented this requirement in Frontiers in Artificial Intelligence. A 2020 field experiment revealed that state legislators responded to AI-generated advocacy letters at rates only 1.9 percentage points lower than human-written letters (15.4% vs 17.3%). Senator Michael Bennet's office announced that proposed legislation in the United States, such as the REAL Political Ads Act, seeks to legally mandate disclaimers on AI-generated political content. The Journal of Democracy underscored the necessity for agencies to adopt automated auditing and machine-readable provenance tracking to verify public input, given AI's ability to flood regulatory comment sections with millions of synthetic inputs. Major technology firms are also deploying technical solutions like embedded watermarks and metadata indicators to establish content origin, as observed by Guillén Yparrea and Iván Miguel García López, and Senator Michael Bennet's office.
The Liar's Dividend Shields Officials
The NAB found that only 26% of respondents trust information produced by AI, with 68% considering it untrustworthy. This widespread public distrust in AI-generated information allows officials to exploit the "liar's dividend," dismissing authentic criticism as "AI noise," even as new regulations emerge. The Brennan Center and Kaylyn Jackson Schiff, Daniel S. Schiff, and Natália S. Bueno in the American Political Science Review examined how this widespread skepticism offers an epistemic shield for politicians, enabling them to dismiss authentic evidence or damaging criticism as "AI noise" or "deepfakes." Kaylyn Jackson Schiff, Daniel S. Schiff, and Natália S. Bueno affirmed that experimental studies confirm "false claims of misinformation significantly increase politician support, particularly against text-based scandals." The Journal of Democracy and Kaylyn Jackson Schiff, Daniel S. Schiff, and Natália S. Bueno reported that institutions respond to AI-generated content at rates statistically indistinguishable from human-written input, suggesting that this classification can enable evasion without necessarily requiring immediate structural changes.
Limitless AI Comments Flood E-Rulemaking
The Journal of Democracy described the challenge regulatory bodies face: limitless unique comments flooding e-rulemaking platforms, making it difficult to ascertain genuine public input. The Journal of Democracy documented how advanced AI easily overcomes older detection methods, such as identifying duplicate text, which previously caught bot-driven astroturfing like the 2017 FCC net neutrality comment flood where 94% of comments were not unique. As a result, AI-generated synthetic sentiment is mechanistically decoupling institutional responsiveness from genuine public will, as existing filtering mechanisms struggle to preserve meaningful democratic accountability. LI Yan in Humanities and Social Sciences Communications, Stanford HAI, and the Journal of Democracy explained that generative AI can produce vast quantities of coherent, human-like content rapidly and cheaply, blurring the line between authentic and inauthentic input.
Fan Li and Ya Yang Study: Labels Fail
A 2024 study by Fan Li and Ya Yang in JMIR Formative Research concluded that while labels help users distinguish AI-generated content from human-created content, they minimally affect perceived accuracy, message credibility, or sharing intention. Mandatory provenance labels and watermarks for AI-generated content have not effectively rebuilt public trust, and their inconsistent enforcement further erodes institutional authority. Fan Li and Ya Yang also observed that, paradoxically, labeling can even slightly enhance sharing intention and perceived accuracy for misinformation in certain contexts. Senator Michael Bennet's office pointed out that inconsistent enforcement, where measures are often voluntary, inconspicuous, easily removed, or rely on self-disclosure, fuels skepticism rather than resolving ambiguity. Fan Li and Ya Yang posited that the "implied truth effect" suggests the absence of a warning, due to inconsistent application, can inadvertently imply veracity, leaving the public uncertain about what to trust. The Journal of Democracy and ODI highlighted how this dynamic contributes to a broader crisis where institutional authority is undermined by blanket skepticism toward all information.
Mandatory Provenance Standards and Auditing
The Journal of Democracy, Guillén Yparrea and Iván Miguel García López, and Alexander Romanishyn, Olena Malytska, and Vitaliy Goncharuk emphasized that governments and regulatory bodies must prioritize the development and mandatory adoption of machine-readable provenance standards and automated auditing systems to verify the authenticity of public input. Classifying AI-generated content as propaganda necessitates a dual approach for institutions: implementing thorough technical and regulatory frameworks for accountability and simultaneously addressing the erosion of public trust. Kaylyn Jackson Schiff, Daniel S. Schiff, and Natália S. Bueno cautioned that without these structural changes, the "liar's dividend" will continue to undermine institutional authority, making it increasingly difficult for genuine public will to influence governance.
Comments ()