Digital technologies are dual-use. The same systems that erode democratic resilience through misinformation and manipulation can strengthen it — through crisis response, accessibility, participation, and transparency. The question is not whether to embrace them, but how to govern them by democratic values.
Democratic resilience in the digital age depends not only on whether information is true or false, but on how contemporary platforms sort, rank, and amplify information before citizens ever see it.
The strongest evidence does not support a simple claim that recommendation algorithms alone cause polarization everywhere. Instead, systems shape exposure via engagement proxies, and downstream effects vary by platform, country, period, and method.
Generative AI changes the risk landscape by lowering the cost of persuasive falsehoods and enabling the liar's dividend — real evidence dismissed as fake, simply because convincing fakes exist.
Platform design has become part of democratic infrastructure — not a purely commercial question.
Democratic resilience — the institutional and societal capacity of a democracy to absorb shocks, adapt under pressure, and preserve competitive elections, pluralistic debate, accountability, and citizen trust. Comparative work distinguishes onset resilience (avoiding autocratization) from breakdown resilience (reversing erosion once it has begun).
In technology policy, the relevant "shock" is not a single tool but a shifting information architecture: recommenders decide what is visible; generative AI expands the supply of synthetic content; platform governance rules set what is removed or monetized; election technology shapes both convenience and confidence.
The institutional and societal capacity of a democracy to absorb shocks, adapt under pressure, and preserve competitive elections, pluralistic debate, accountability, and citizen trust.
Rank posts, videos, and accounts using engagement proxies — determining what is repeatedly seen.
Foundation models producing synthetic text, image, audio and video — lowering the cost of persuasive falsehoods at scale.
What is removed, labeled, demoted, monetized — and who decides. Includes recommender-system rules, terms of service and appeal mechanisms.
From voter-registration databases and ballot-marking devices to online political ads, i-voting and deepfake defenses.
The reputational benefit bad actors extract from the mere existence of convincing deepfakes: dismissing genuine evidence as fake.
State-provided digital rails — identity, payments, data exchange — that citizens and services rely on (e.g. Estonia's e-ID and X-Road).
Under the EU Digital Services Act, harms very large platforms must assess and mitigate: illegal content, fundamental-rights impact, civic discourse, elections, public security, and public health.
Networks of accounts operating together to manipulate public debate while misrepresenting who is behind them.
Cryptographic or perceptual signals attached to media that let downstream viewers verify origin and detect edits (e.g. C2PA).
Coordinated, largely non-illegal but manipulative behavior by foreign actors that threatens democratic values, processes and institutions.
Because recommenders run continuously and at scale, even small ranking preferences reshape the visibility of political speech, the speed of rumor propagation, and the perceived boundaries of mainstream opinion.
The evidence since 2016 points to a more precise claim than "algorithms radicalize." Recommendation systems are best understood as risk multipliers: they magnify existing cleavages, partisan ecosystems, and coordinated disinformation — but their effects depend on user demand, network composition, and local media systems.
The regulatory and empirical landscape has moved quickly. Click a marker to expand.
Structural regimes shift from adoption to enforcement; new research pins causal effects of generative AI.
Governments start moving from voluntary self-regulation to structured intermediary obligations.
COVID-19 turns platform moderation into a public-health question and accelerates election-tech scrutiny.
Empirical work moves from anecdote to randomized experiment.
The EU seals its two-track digital rulebook and Musk's Twitter acquisition reshapes the largest political platform.
Voluntary AI risk frameworks land; Estonia proves an internet-majority election is possible.
A record election year meets the first binding AI regulation and the first landmark synthetic-media enforcement.
Structural regimes shift from adoption to enforcement; new research pins causal effects of generative AI.
AI transparency, provenance and researcher-access rules approach full application.
Six overlapping mechanisms move platforms from mere curation toward democratic consequence.
Ranking determines which narratives are repeatedly seen. Early performance advantages compound: visibility begets engagement, engagement begets visibility.
Google Search research shows partisanship correlates less with what search surfaces than with what users click — algorithms and users are complements, not substitutes.
Twitter amplification favored the mainstream right in 6/7 countries — without preferring the farthest fringe. Attention can still shift asymmetrically across blocs.
YouTube sock-puppet audits found ideologically biased and increasingly radical recommendations for some profiles; newer naturalistic studies find limited short-term attitude change.
Meta election experiments changed exposure but showed limited attitude shifts — while highlighting how hard it is to study platforms whose systems change mid-experiment.
Synthetic text, audio, and video slash the cost of manipulation. Even when exposed, fakes persist — and their existence lets bad actors deny authentic recordings.
Twelve cases across North America, Europe, Asia, Latin America, MENA and Africa — some illustrate democratic risks from algorithmic and synthetic-media systems, others show how the same technologies can strengthen democratic capacity.
Secure digital identity underpins nearly every interaction between Estonians and their state — including voting. In the 2023 parliamentary election, more than half of all ballots were cast online.
vTaiwan and the Join platform channel deliberation between citizens, civil society, experts, and government — an alternative to a purely defensive information posture.
Forecasting and analytics support earthquake, typhoon, and extreme-weather preparedness, sustaining institutional legitimacy through effective crisis response.
High-resolution digital twins of Earth systems let policymakers stress-test climate responses before crises arrive.
NOAA-supported community campaigns combine satellite imagery, sensors, and volunteer data collection to map heat exposure block-by-block, revealing environmental inequality.
The world's largest democracy has embraced digital campaigning for reach — while facing scaled synthetic media, encrypted misinformation, and opaque political targeting.
A New Hampshire primary robocall using an AI clone of President Biden's voice discouraged participation. Regulators finalized a $6M fine — a landmark synthetic-media election-integrity action.
The Public Official Election Act was amended to ban AI-generated deepfakes in campaigns within 90 days of voting, with criminal penalties.
Clema's ClemaCore is a solid-state, battery-powered cooling vest designed for industrial and military crews working in 120°F+ heat — an example of how private climate-adaptation technology can shore up the state capacity democracies need during increasingly severe heat waves.
Brazil's 2022 general election showed how encrypted messaging and synthetic media can simultaneously threaten and strengthen democratic resilience. The Superior Electoral Court (TSE) partnered with platforms to fact-check viral content, while WhatsApp remained the primary vector for coordinated disinformation.
Kenya's 2022 general election was preceded by a surge in ethnically targeted hate speech on Facebook, TikTok and Twitter. The National Cohesion and Integration Commission (NCIC) warned that platform moderation teams lacked sufficient Swahili and local-language capacity to catch coded incitement.
Tunisia's 2011 revolution was organised on Facebook and blogs, earning the region the label 'Facebook revolution.' A decade later, Decree 54 (2022) criminalised the spread of 'false news' online, with dozens of journalists and activists prosecuted under broad language that international watchdogs warn chills legitimate speech.
Nigeria's 2023 general election saw an unprecedented scale of fact-checking collaboration. Organisations including Dubawa, Africa Check and the Centre for Democracy and Development ran real-time verification of candidate claims, viral images and manipulated video across Hausa, Yoruba, Igbo and English.
Ukraine's Diia app — launched in 2020 — became a wartime backbone after February 2022, delivering documents, IDP registration, damage claims and even bond purchases to citizens whose physical institutions were under attack.
Mexico's 2024 elections — the largest in the country's history — saw AI-generated audio and video of candidates circulate on TikTok, X and WhatsApp. The National Electoral Institute (INE) issued the region's first binding guidance on synthetic campaign content.
After the September 2022 death of Mahsa Amini in police custody, Iranian authorities imposed the country's most severe internet restrictions on record — throttling mobile data, blocking Instagram and WhatsApp, and disconnecting entire provinces for days at a time.
Secure digital identity underpins nearly every interaction between Estonians and their state — including voting. In the 2023 parliamentary election, more than half of all ballots were cast online.
An evidence map, not a scorecard. Different platforms, methods, and outcome measures reveal different parts of the same problem.
~2M daily active Twitter accounts
Mainstream right amplified more than mainstream left in 6 of 7 countries; extremes not amplified more than centrists.
Cross-national asymmetry — but warns against equating amplification with automatic extremist favoritism.
Facebook & Instagram, 2020 cycle (n≈23,000)
Removing algorithmic ranking changed exposure but showed limited attitude effects in-study.
Cautions against simple causal claims; strengthens the case for independent data access.
Facebook users (n≈23,000)
Like-minded sources common; reducing exposure to them did not measurably shift polarization in-study.
Content exposure ≠ persuasion.
Facebook & Instagram deactivation, 6 wks before 2020 election (n≈35,000)
Deactivating Facebook modestly reduced polarization and news knowledge; Instagram effects smaller.
Platform-level effects exist but are heterogeneous by service and user.
YouTube-like recommendation environment (n≈9,000)
Recommendation slant changed viewing behavior a lot; produced little short-term opinion change.
Separate behavioral and attitudinal effects analytically.
10.1M Facebook users with self-reported ideology
Individual choice reduced cross-cutting content more than the News Feed algorithm did.
Early quantification of the algorithm-vs-user contribution to echo chambers.
GPT-4 vs human debaters with personal data (n=900)
GPT-4 with limited personal data was 81.7% more persuasive than a human opponent.
First causal evidence microtargeted LLMs can out-persuade humans.
100,000 bot accounts
Recommendations ideologically biased and increasingly radical, especially for right-leaning profiles.
Clearest audit-based warning on drift pathways.
350,000+ YouTube videos
Hateful videos common in search and first-level recommendations; exposure varied by gender and religion.
Shows identity-sensitive harms beyond Western platforms.
170,000+ videos across 5 topics
Personalization can push users into misinformation-heavy filter bubbles on health & science topics.
Audit method translates outside politics — same drift, different domain.
330,000 videos, 2M recommendations
Users migrated from Intellectual Dark Web → alt-lite → alt-right channels via recommendations & comments.
Documented radicalization pipeline before YouTube's 2019 policy changes.
Google Search panel, 2018 & 2020 elections (n≈1,000)
Partisan/unreliable news exposure tied more to users' engagement choices than to what search surfaced.
Design and citizen capability must be regulated together.
Web-tracking panels (n≈4,500)
Untrustworthy-news exposure concentrated in a small minority of heavy consumers.
Reframes 'misinformation crisis' as a distributional problem, not a mass one.
n≈95,000
40% trust most news most of the time; direct visits to news sites decline as platform intermediation rises.
Sets the trust substrate on which recommenders and AI operate.
n≈3,000, fact-checking behaviour
Using search engines to evaluate news increased belief in misinformation for low-quality articles.
Media-literacy prescriptions ('just search it') can backfire under data-void conditions.
n≈2,500 WhatsApp users during 2018 election
Pro-Bolsonaro disinformation was shared more than pro-Haddad content on WhatsApp; forwarding behaviour concentrated among politically engaged users.
First large-sample evidence that encrypted messaging is a major political information vector outside the US and Europe.
Cross-cultural comparison of filter bubbles
Filter bubbles vary significantly by cultural and political context; democratic and authoritarian contexts show different echo-chamber structures.
Cautions against exporting US-centric platform conclusions to Latin America, MENA and Africa.
Focus groups and surveys on news consumption
Users in the Global South rely heavily on WhatsApp and Facebook for news; trust in platforms often exceeds trust in legacy media.
Platform governance in these markets must account for WhatsApp-centric information architectures and lower legacy-media trust.
Expert-coded panel, 1789–2024
Average world citizen's level of democracy back to 1985 levels; 72% of the world lives in autocracies.
Places platform/AI risk inside a wider autocratization trend.
Internet population coverage ~87%
Global internet freedom declined for the 14th consecutive year; generative AI used to distort information in 16+ national elections.
AI-manipulation is now a governance-relevant metric, not hypothetical.
Expert scoring on 60 indicators
Global score fell to 5.17 — the lowest since the index began in 2006; only 6.6% of people live in a 'full democracy'.
Second index converging on the same erosion signal.
Composite indicators across 5 categories
Half of countries with democratic performance measured declined on at least one democratic-performance category in the last 5 years.
Confirms breadth of decline across representation, rights, rule of law, participation.
TikTok videos reported during election period
Platform failed to act on 90% of reported hate-speech videos in Swahili and English.
Highlights the multilingual content-moderation gap in African electoral contexts.
YouTube recommendation audit during 2022 election
Right-leaning and far-right channels received more recommendation traffic than left-leaning equivalents.
Shows ideological asymmetry in recommender systems extends beyond the US and Europe.
Governments are converging on a shared menu — transparency, risk management, election-specific rules, intermediary accountability, and targeted enforcement — even where their regulatory philosophies differ.
Voluntary risk-management framework for trustworthy AI, with a 2024 profile for generative AI.
Useful for audits and organizational controls; not a binding platform-democracy regime.
Accessible voting systems; privacy-independent participation; usability, auditability, integrity.
Election tech can improve access while embedding verification.
AI-generated voices in robocalls treated as unlawful; $6M fine in the Biden voice-clone case.
Targeted election-integrity response to synthetic media.
Systemic-risk regime for VLOPs; recommender transparency; non-profiling option; Commission investigative powers.
The most developed structural framework for platform accountability.
Risk-based AI regulation; transparency for AI-generated content under Art. 50 (key obligations from Aug 2026).
Directly relevant to deepfakes, provenance, and disclosure.
Voluntary commitments on demonetization, political-ad transparency, research access.
Useful soft-law layer; effectiveness evidence mixed.
Correction directions, disabling powers, and codes for political-ad transparency and credible sources.
Strong administrative toolkit; criticized for broad executive discretion.
Restrictions on foreign online election ads; ban on manipulated content misrepresenting candidates during elections.
Election-specific hardening against foreign influence & synthetic media.
Criminalizes election-related acts funded/instructed by hostile foreign infiltration sources.
Sovereignty-centered foreign-interference model.
Ban on AI-generated deepfakes in campaigns within 90 days of voting, with criminal penalties.
Focused response to synthetic-media electoral manipulation.
Legislative proposal requiring platform transparency, AI-content labelling, and due-process rules for content removal; stalled by platform opposition and free-speech concerns.
Would create the most comprehensive platform-regulation regime in the region if enacted.
Binding electoral guidance requiring parties and candidates to label AI-generated material and banning deepfakes impersonating opponents.
Shows how an election-management body can move faster than statute on generative AI.
Constitutional data-protection right (2018) plus a 2023 executive-branch working group coordinating platform, academic and press response to electoral disinformation.
Combines rights-based data protection with a soft-law coordination model instead of speech criminalisation.
Communications regulator (CRC) content-moderation transparency guidance; National Electoral Council rules on online political advertising and disclosure.
Regulator-led transparency model rather than a single omnibus statute.
Criminalises false publication, cyber-harassment and fake news; amended to narrow scope after court challenges.
Illustrates the tension between election-integrity goals and overbroad speech restrictions that chill dissent.
Regulates online political advertising and requires disclosure of sponsors; complemented by independent fact-checking coalitions (Dubawa, Africa Check).
Hybrid model: statutory rules on ads plus civic verification infrastructure rather than top-down removal.
POPIA (2020) enforces data-protection rights against platforms; the 2019 Films and Publications amendments create offences for non-consensual intimate imagery and 'harmful' online content.
One of the region's most developed rights-based frameworks; enforcement capacity remains the constraint.
Establishes the Cyber Security Authority, mandates critical-information-infrastructure protection, and pairs with National Communications Authority guidance on election-period platform conduct.
Institutional-capacity approach: build a regulator first, then layer election-specific rules on top.
Criminalises the creation and spread of 'false news' with up to five years in prison; applied against journalists, activists and opposition figures.
Demonstrates how counter-disinformation laws can be repurposed to suppress civic discourse.
State-led AI industrial strategy paired with a cybercrimes law that criminalises 'rumours' and content 'harmful to public order', with penalties up to life imprisonment for some offences.
Illustrates a state-capacity model that pairs heavy AI investment with tight speech control — the inverse of the EU's rights-based framing.
Broadly criminalises online 'false news', 'contempt of religion' and content 'undermining national unity', with prison terms and platform-blocking powers.
A recent regional example of the counter-disinformation frame being used to narrow civic space; drew criticism from UN Special Rapporteurs.
One of the region's earliest data-protection regimes (CNDP regulator) alongside a controversial (later shelved) 2020 draft criminalising 'fake news' on social media.
Shows the tension between a functioning data-protection authority and periodic legislative attempts to restrict online speech.
Statutory basis for internet voting in Riigikogu Election Act; independent audit and source-code publication after the 2023 election where >50% of ballots were cast online.
Most mature legal-technical framework for binding online voting in a democracy.
2023–24 amendments require online political-ad disclosure; the KRRiT broadcasting regulator has expanded remit over on-demand and platform content, alongside the National Cybersecurity Strategy 2019–24.
Post-2023 government has pushed reforms to align with the EU DSA and rebuild public-media independence.
Law on Public Electronic Services underpins the Diia app; a National Security and Defence Council body coordinates wartime rapid response to Russian influence operations.
Live example of digital public infrastructure and information defence operating under wartime conditions.
After the Constitutional Court annulled the 2024 presidential first round over undisclosed TikTok campaign spending, the Permanent Electoral Authority (AEP) and audiovisual regulator (CNA) tightened rules on online political advertising and platform reporting.
First EU case where an election was annulled over social-media manipulation — a stress test for DSA enforcement.
Voluntary risk-management framework for trustworthy AI, with a 2024 profile for generative AI.
Useful for audits and organizational controls; not a binding platform-democracy regime.
Accessible voting systems; privacy-independent participation; usability, auditability, integrity.
Election tech can improve access while embedding verification.
AI-generated voices in robocalls treated as unlawful; $6M fine in the Biden voice-clone case.
Targeted election-integrity response to synthetic media.
Systemic-risk regime for VLOPs; recommender transparency; non-profiling option; Commission investigative powers.
The most developed structural framework for platform accountability.
Risk-based AI regulation; transparency for AI-generated content under Art. 50 (key obligations from Aug 2026).
Directly relevant to deepfakes, provenance, and disclosure.
Voluntary commitments on demonetization, political-ad transparency, research access.
Useful soft-law layer; effectiveness evidence mixed.
Correction directions, disabling powers, and codes for political-ad transparency and credible sources.
Strong administrative toolkit; criticized for broad executive discretion.
Restrictions on foreign online election ads; ban on manipulated content misrepresenting candidates during elections.
Election-specific hardening against foreign influence & synthetic media.
Criminalizes election-related acts funded/instructed by hostile foreign infiltration sources.
Sovereignty-centered foreign-interference model.
Ban on AI-generated deepfakes in campaigns within 90 days of voting, with criminal penalties.
Focused response to synthetic-media electoral manipulation.
Legislative proposal requiring platform transparency, AI-content labelling, and due-process rules for content removal; stalled by platform opposition and free-speech concerns.
Would create the most comprehensive platform-regulation regime in the region if enacted.
Binding electoral guidance requiring parties and candidates to label AI-generated material and banning deepfakes impersonating opponents.
Shows how an election-management body can move faster than statute on generative AI.
Constitutional data-protection right (2018) plus a 2023 executive-branch working group coordinating platform, academic and press response to electoral disinformation.
Combines rights-based data protection with a soft-law coordination model instead of speech criminalisation.
Communications regulator (CRC) content-moderation transparency guidance; National Electoral Council rules on online political advertising and disclosure.
Regulator-led transparency model rather than a single omnibus statute.
Criminalises false publication, cyber-harassment and fake news; amended to narrow scope after court challenges.
Illustrates the tension between election-integrity goals and overbroad speech restrictions that chill dissent.
Regulates online political advertising and requires disclosure of sponsors; complemented by independent fact-checking coalitions (Dubawa, Africa Check).
Hybrid model: statutory rules on ads plus civic verification infrastructure rather than top-down removal.
POPIA (2020) enforces data-protection rights against platforms; the 2019 Films and Publications amendments create offences for non-consensual intimate imagery and 'harmful' online content.
One of the region's most developed rights-based frameworks; enforcement capacity remains the constraint.
Establishes the Cyber Security Authority, mandates critical-information-infrastructure protection, and pairs with National Communications Authority guidance on election-period platform conduct.
Institutional-capacity approach: build a regulator first, then layer election-specific rules on top.
Criminalises the creation and spread of 'false news' with up to five years in prison; applied against journalists, activists and opposition figures.
Demonstrates how counter-disinformation laws can be repurposed to suppress civic discourse.
State-led AI industrial strategy paired with a cybercrimes law that criminalises 'rumours' and content 'harmful to public order', with penalties up to life imprisonment for some offences.
Illustrates a state-capacity model that pairs heavy AI investment with tight speech control — the inverse of the EU's rights-based framing.
Broadly criminalises online 'false news', 'contempt of religion' and content 'undermining national unity', with prison terms and platform-blocking powers.
A recent regional example of the counter-disinformation frame being used to narrow civic space; drew criticism from UN Special Rapporteurs.
One of the region's earliest data-protection regimes (CNDP regulator) alongside a controversial (later shelved) 2020 draft criminalising 'fake news' on social media.
Shows the tension between a functioning data-protection authority and periodic legislative attempts to restrict online speech.
Statutory basis for internet voting in Riigikogu Election Act; independent audit and source-code publication after the 2023 election where >50% of ballots were cast online.
Most mature legal-technical framework for binding online voting in a democracy.
2023–24 amendments require online political-ad disclosure; the KRRiT broadcasting regulator has expanded remit over on-demand and platform content, alongside the National Cybersecurity Strategy 2019–24.
Post-2023 government has pushed reforms to align with the EU DSA and rebuild public-media independence.
Law on Public Electronic Services underpins the Diia app; a National Security and Defence Council body coordinates wartime rapid response to Russian influence operations.
Live example of digital public infrastructure and information defence operating under wartime conditions.
After the Constitutional Court annulled the 2024 presidential first round over undisclosed TikTok campaign spending, the Permanent Electoral Authority (AEP) and audiovisual regulator (CNA) tightened rules on online political advertising and platform reporting.
First EU case where an election was annulled over social-media manipulation — a stress test for DSA enforcement.
Five short scenarios drawn from the brief. Guess whether each describes a genuine event or an AI-manipulated one — then read what actually happened.
A robocall using President Biden's voice tells New Hampshire voters to 'save your vote' and skip the primary.
Seven synthesis points across the case studies, comparative evidence and policy landscape — what a democratic-resilience roundtable should treat as the working consensus.
Cross-national experiments (Huszár 2021) confirm that recommender systems asymmetrically amplify political content — but effects vary by country, platform, and period, and rarely translate directly into attitude change (Guess 2023, Nyhan 2023, Liu 2025).
Removing algorithmic ranking or reducing like-minded exposure changes what people see without reliably shifting what they believe in short study windows. Behavioural and attitudinal effects must be separated analytically.
Salvi et al. (2024) show GPT-4 with limited personal data out-persuades humans on political topics by 81.7%. Combined with cheap voice cloning (Biden robocall) this changes the electoral-manipulation cost curve.
It is no longer hypothetical: candidates and officials routinely dismiss authentic recordings as AI-generated. Provenance, watermarking (C2PA) and disclosure regimes matter for both directions.
Estonia (DPI + i-voting), Taiwan (participation + Anti-Infiltration Act), Japan (crisis AI), the EU (DSA + AI Act), Brazil (TSE fact-checking), Nigeria (civic verification coalitions) and firms like Clema (heat resilience) show capacity is built across identity, participation, adaptation and enforcement layers — never one lever alone.
Transparency, systemic-risk management, election-specific synthetic-media rules, codes of practice, and targeted enforcement now appear across the US, EU, Asia, Latin America, MENA and Africa — with sharp differences in enforceability and executive discretion.
V-Dem 2025 puts the average citizen's level of democracy at 1985 levels; Freedom House flags 14 straight years of declining internet freedom and AI-manipulation in 16+ national elections. Platform reform sits inside a wider autocratization trend.
Digital technology is neither the disease nor the cure of democratic decline. It is a force multiplier — of the institutions, participation habits and adversaries a democracy already has.
Hover — or, on a phone, drag your finger across a card — and a lens follows your cursor, revealing how the same civic behaviour looks for a generation that grew up in digital third spaces.
Voters historically gathered in cafés, pubs, churches and union halls — 'third places' where face-to-face political conversation and organising happened. Many 19th- and 20th-century political movements were centred on them.
Evening broadcast, morning paper, trusted editorial gatekeepers deciding what is on the front page.
A rally, a leaflet, a letter to the editor. Sincerity is the default register; length signals seriousness.
'Seeing is believing.' Video was the gold standard of evidence in journalism and courts alike.
You vote where you were raised, in person, on election day. Civic duty framed as ritual.
Inherited from family, reinforced by union, church or neighborhood; stable across a lifetime.
The lens is illustrative, not a survey instrument — pointers to well-documented generational shifts in political sociology and media use.
Strong studies use short windows or platform cooperation, and platforms change ranking systems mid-study.
The most rigorous experimental evidence still comes disproportionately from the US and English-language platforms. Latin America, MENA and Africa are under-represented despite being high-risk, high-growth digital markets.
Outside the DSA, researchers depend on temporary access arrangements, disclosures, or audits.
This brief introduces synthetic media and election tech but does not yet fully analyze microtargeting, ad markets, and LLM persuasion research.
No single intervention is sufficient. The evidence supports a layered democratic-resilience approach — reduce technological risks through transparency, auditing, and accountability, while investing in technologies that strengthen democratic capacity.
Prepared with the support of the U.S. Department of State and the Community of Democracies as part of ongoing work on democratic resilience in the digital age. Views expressed are the authors' own and do not necessarily reflect those of the supporting institutions.