Report 28 — Synthetic Consensus, Deepfakes, and the Epistemic War Around Uprisings

Full report

On this page

Research question

What happens to political legitimacy when machine intelligence can manufacture convincing evidence, millions of tailored messages, synthetic supporters, and entire apparent social movements? How can a transition distinguish genuine public consent from generated consensus without creating a censorship regime?

Executive finding

The AI-era political crisis is not only a fight over what happened. It is a fight over whether shared verification is still possible. Freedom House documented generative-AI use in election campaigns and information manipulation, while V-Dem reports a close association among government disinformation, polarization, and autocratization. International IDEA’s recent election work likewise treats information integrity as a core democratic resilience issue.

The danger is symmetrical: authorities can fabricate public support; opposition actors can fabricate atrocities or endorsements; foreign actors can manufacture both sides; and genuine evidence can be dismissed as artificial.

1. Synthetic consensus changes the meaning of “public opinion”

Political systems already use polling, focus groups, social media trends, rallies, petitions, comments, and news coverage as evidence of public sentiment. Machine intelligence can contaminate every one of these proxies.

Synthetic personas can create:

• apparent grassroots outrage;
• fake defections;
• fake endorsements;
• fake whistleblowers;
• fake victims;
• fake crowds in images;
• fake recordings of political leaders;
• artificial comment majorities;
• mass personalized persuasion.

This means future uprisings may begin in an environment where neither government nor opposition can easily prove that apparent public sentiment is real.

2. The liar’s dividend will protect real misconduct

Once citizens know deepfakes exist, any authentic recording can be denied. This is politically useful to actors with something to hide.

The most dangerous moment is a transition crisis: an order to use force, a secret donor conversation, a corrupt bargain, a military refusal, or an admission of election manipulation may be genuine, yet audiences can be told that AI produced it.

Prediction: provenance becomes constitutional infrastructure.

3. The return of institutions as trust anchors

Paradoxically, AI may make trusted institutions more important, not less. When individual pieces of media become cheap to fabricate, verification depends on networks of independent corroboration:

• courts;
• election officials;
• professional newsrooms;
• public archives;
• universities and forensic labs;
• international observers;
• civil society monitors;
• authenticated government records;
• cryptographic provenance systems.

The political battle will therefore include efforts to destroy trust in these institutions before a crisis arrives.

4. Prediction: every side will accuse the others of being “bots”

Dehumanization can take a new form. Instead of calling opponents traitors or extremists, actors can dismiss them as fake accounts, paid operatives, foreign agents, or AI-generated identities.

Some accusations will be true. Others will become a convenient way to deny that real citizens hold disliked opinions.

A democracy cannot solve synthetic participation by presuming that unpopular speech is artificial.

5. Machine intelligence lowers the cost of authoritarian narrative control

An authoritarian information system no longer needs a single propaganda line. AI can generate different narratives for different audiences:

• nationalism for one group;
• economic stability for another;
• religious defense for another;
• fear of disorder for another;
• anti-corruption rhetoric for another;
• claims of foreign conspiracy for another.

Contradictions matter less when each audience receives its own story.

This is a transition from mass propaganda to personalized ideological containment.

6. Opposition movements face the same temptation

A movement fighting a manipulative state may justify its own fabricated media as necessary counter-propaganda. That is strategically corrosive.

Once a movement normalizes synthetic evidence, it destroys the evidentiary system it will need to govern afterward.

The AMIANISM principle “PRESERVE THE SOURCE” therefore becomes directly political: preserve originals, transformations, dates, authorship, model involvement, and chain of custody.

7. Post-transition accountability requires an evidence architecture

If a society wants to determine who actually subverted democratic processes, it needs more than accusations. It needs records capable of surviving adversarial review.

A transition-era evidence system should distinguish:

• original source material;
• authenticated copies;
• AI-enhanced or translated material;
• synthetic reconstructions;
• anonymous allegations;
• adjudicated findings;
• contested claims.

This is especially important before imposing disqualification from office. A political enemy should not be excluded because a viral clip appears incriminating.

8. Preventing the return of anti-democratic networks through information rules

The durable tools are structural:

• rapid disclosure of political advertising sponsors;
• beneficial ownership of organizations buying political media;
• public archives of paid political ads;
• disclosure when campaigns use synthetic candidate likenesses;
• independent election-information channels;
• penalties for fraudulent impersonation rather than broad bans on political speech;
• media ownership transparency;
• protection for researchers and journalists studying influence operations;
• public access to verified transition records.

These rules apply to all political ecosystems, including Democratic and Republican organizations in a U.S. context.

9. Why censorship is a dangerous answer

A new government may be tempted to designate the previous coalition’s narratives as “disinformation” and suppress them. This can quickly reproduce the old problem under new management.

Democratic resilience requires narrow rules tied to provable conduct—fraud, impersonation, unlawful foreign financing, threats, covert coordinated activity—not broad state power to declare political interpretation false.

10. Forecast: truth will become slower than virality

AI systems can generate content instantly. Verification remains slower because it depends on evidence and independent checking.

This produces an asymmetry: falsehood can dominate the first hours of a crisis.

The institutions most likely to survive this environment are those that build preexisting trust before the crisis, publish transparent methods, correct errors, and preserve verifiable records.

11. Strongest conclusion

Machine intelligence turns legitimacy into a provenance problem. The central struggle of an AI-era uprising may be whether a society can still establish a common factual record. A democratic transition that cannot prove what happened will either collapse into revenge and rumor or hand enormous censorship power to whoever claims to be the truth authority. The alternative is source integrity, independent verification, transparent political-media financing, and due process before sanction.

Sources

1. Freedom House, “Election Meddling and a Growing Assault on Free Expression Online…” (2024): https://freedomhouse.org/article/new-report-election-meddling-and-growing-assault-free-expression-online-led-14th
2. V-Dem, Democracy Report 2025: https://www.v-dem.net/documents/54/v-dem_dr_2025_lowres_v1.pdf
3. International IDEA, “Safeguarding Democracy: Elections, Information Integrity and Artificial Intelligence” (2025): https://www.idea.int/news/safeguarding-democracy-eu-development-nexus-elections-information-integrity-and-artificial
4. UNESCO, “Freedom of Expression, AI and Elections” (2024): https://www.unesco.org/en/articles/8000-enroll-new-course-freedom-expression-ai-and-elections
5. AMIANISM source principle: PRESERVE THE SOURCE. LABEL THE DERIVATIVE.