Report 27 — AI Surveillance, Digital Repression, and Mass Mobilization

Full report

On this page

Research question

How will machine intelligence alter the balance between mass civic mobilization and state repression, and what institutional safeguards reduce the risk that democratic governments inherit and normalize an authoritarian surveillance architecture after a transition?

Executive finding

The defining change is not simply “more surveillance.” It is integration: cameras, biometric systems, communications metadata, social media, financial records, public databases, and AI classification can be fused into near-real-time profiles. Carnegie’s 2025 work describes a shift from blunt censorship toward targeted digital repression using facial recognition, spyware, metadata analysis, social listening, and AI-driven profiling. Freedom House likewise documents expanding surveillance and censorship around protests and elections.

This creates a state that may be able to impose consequences on selected participants without visibly repressing everyone. That makes accountability harder because repression can be individualized and bureaucratized.

1. From crowd control to network control

Traditional protest policing focuses on places: the square, march route, union hall, campus, government building. Machine intelligence enables authorities to focus on relationships.

A network model can infer:

• recurring organizers;
• communication hubs;
• journalists and legal observers;
• financial and logistical dependencies;
• influential accounts;
• links between unions, civic groups, parties, and donors;
• geographic patterns of mobilization.

The political implication is profound: a movement can be weakened without banning the movement as such. Pressure can be applied to nodes that make participation sustainable.

2. The likely rise of “administrative repression”

Future repression may increasingly arrive as ordinary procedure rather than overt emergency rule:

• licensing and permit denials;
• employment or contracting consequences;
• tax and compliance scrutiny;
• immigration or travel consequences;
• risk scores;
• platform demonetization or throttling;
• automated fraud or security flags;
• banking or payment restrictions;
• algorithmic watchlists.

Each action can appear technically justified in isolation. Together they can form a Cage.

This is especially dangerous because it diffuses responsibility. No one official needs to say “silence this dissident.” A chain of models and procedures can produce the same result.

3. Prediction: selective repression often replaces spectacular repression

Where institutions remain partly democratic, leaders face reputational and legal costs for mass violence. Machine intelligence offers a quieter alternative: identify high-impact participants and increase their personal costs.

This does not eliminate violence, especially where states are already repressive. But it can reduce the need for visible coercion.

The result is a legitimacy challenge: citizens may experience the state as repressive while every agency insists it merely followed neutral procedure.

4. Automated suspicion can become self-reinforcing

AI systems trained on records created by prior policing can reproduce prior enforcement patterns. If a neighborhood, political community, or organization was heavily monitored in the past, more data exist about it; the model can then interpret the abundance of records as evidence of greater risk.

In transition, simply keeping the old model can therefore preserve the old political hierarchy without a conscious decision to do so.

Prediction: democratic recovery will increasingly require model transition, not only personnel transition.

5. What happens when an old regime falls?

Surveillance systems create a dangerous inheritance. The successor government acquires:

• watchlists;
• informant records;
• biometric databases;
• model outputs;
• social graphs;
• communications intercept archives;
• protest intelligence;
• political-risk labels.

The temptation to use them against yesterday’s rulers will be enormous.

That temptation is the democratic test.

If the new coalition simply points the same machinery at its enemies, it confirms the old premise that power has a right to know everything about everyone.

6. A democratic transition needs a “surveillance inventory”

One of the first institutional reform tasks should be a lawful inventory of state surveillance capability:

• what systems exist;
• which agencies operate them;
• what data they ingest;
• what legal authority supports them;
• retention periods;
• vendor contracts;
• automated decision functions;
• access controls;
• foreign intelligence interfaces;
• audit logs;
• appeal mechanisms.

The objective is not to destroy legitimate investigative capacity. It is to separate legitimate, warrant-based, reviewable security functions from political monitoring and unreviewable population scoring.

7. Preservation before deletion

Paradoxically, immediate destruction of old surveillance files can undermine justice. Records may document abuses, unlawful targeting, chains of command, and victims.

The OHCHR’s transitional-justice guidance emphasizes the importance of records for prosecution, truth processes, vetting, institutional reform, and historical accountability.

Therefore the correct sequence is often:

secure → preserve → independently review → classify lawful access → remedy → then lawfully delete or archive where appropriate.

This prevents both cover-up and permanent operational reuse.

8. How to prevent anti-democratic networks from retaining surveillance power

The focus should be on functions and access, not political identity.

Safeguards include:

• warrants or clearly defined statutory authorization for intrusive surveillance;
• independent judicial and legislative oversight;
• inspector-general powers;
• logs that record who queried which person and why;
• criminal and civil penalties for unauthorized political surveillance;
• mandatory periodic review of watchlists;
• prohibition on using protected political belief alone as a risk factor;
• human appeal for consequential automated decisions;
• procurement transparency and vendor accountability;
• sunset clauses for emergency surveillance authorities.

If individuals from Democratic, Republican, or any other political networks participated in unlawful surveillance or abuse, the relevant criterion should be evidence of conduct—not party membership.

9. The supporter problem

A surveillance regime often depends on many ordinary professionals: engineers, analysts, police officers, contractors, telecom workers, administrators. Collective dismissal would cripple the state and erase expertise.

The democratic alternative is role-sensitive vetting:

• distinguish policy architects from routine staff;
• distinguish knowing abuse from good-faith implementation of lawful duties;
• retain technical competence where integrity is established;
• require new oaths/rules/training tied to constitutional rights;
• create protected channels for internal reporting.

The objective is to turn the coercive apparatus back into a public service.

10. Strongest conclusion

In the machine-intelligence era, repression may become less visible precisely because it becomes more data-driven. The democratic transition problem is therefore not only who controls the police. It is who controls the models, databases, watchlists, vendor systems, and access privileges that define who is suspicious. A legitimate successor government must preserve evidence of abuse while preventing the old surveillance architecture from becoming its own instrument of revenge.

Sources

1. Carnegie Endowment, “Shrinking Civic Space, Digital Funding, and Legitimacy in a Post-Truth Era” (2025): https://carnegieendowment.org/research/2025/07/shrinking-civic-space-digital-funding-and-legitimacy-in-a-post-truth-era
2. Freedom House, Freedom on the Net 2025: https://freedomhouse.org/report/freedom-net/2025
3. OHCHR, Rule-of-Law Tools for Post-Conflict States: Vetting: https://www.ohchr.org/Documents/Publications/RuleoflawVettingen.pdf
4. OHCHR, Rule-of-Law Tools for Post-Conflict States: Archives: https://www.ohchr.org/sites/default/files/Documents/Publications/HR_PUB_14_4_Archives_en.pdf
5. DCAF, Intelligence Accountability: https://www.dcaf.ch/intelligence-accountability