Report 29 — War at Machine Speed: Security Forces, State Capacity, and the Coercive Apparatus During Political Transition

Full report

On this page

Research question

How does machine intelligence change the role of police, military, intelligence services, emergency management, and automated command systems when a political system enters a legitimacy crisis? What happens to these institutions after a transition?

Executive finding

Historical transitions repeatedly turn on the behavior of the coercive apparatus. Machine intelligence does not remove that fact; it changes the operating tempo and diffuses responsibility. AI can accelerate intelligence analysis, targeting support, logistics, communications monitoring, resource allocation, and public-order decision making. This may make security institutions more capable but also more dependent on opaque technical systems.

The core risk is decision acceleration without accountability.

1. The coercive apparatus still decides whether a crisis becomes a transition

Book Three correctly emphasizes that police, military, intelligence services, guards, and rank-and-file personnel may repress, hesitate, split, defect, or refuse. Historical research on revolutions consistently shows that mass numbers alone do not determine outcomes.

Machine intelligence may strengthen centralized command, but it cannot eliminate the human legitimacy problem. Orders still need institutions to execute them, equipment to maintain them, budgets to pay them, and personnel to accept responsibility.

2. Machine systems can make repression more administratively efficient

AI-assisted security systems can help prioritize people, locations, communications, and incidents. In a crisis, this can reduce response times and increase the state’s ability to deploy resources selectively.

That same capacity can be used legitimately—for emergency medicine, de-escalation, traffic management, or protection of infrastructure—or illegitimately for political repression.

Machine is therefore not the moral actor by default. The question is who defines the objective, what constraints exist, and who can appeal.

3. “The model recommended it” becomes a new form of responsibility laundering

A dangerous future pattern is the use of AI to distance humans from coercive decisions:

The system flagged the person. The model rated the event high-risk. The software recommended deployment. The commander followed protocol.

Responsibility becomes distributed across code, vendor, analyst, supervisor, and institution.

A democratic transition must reject the idea that automation dissolves accountability.

4. Prediction: security-force fractures may become data fractures

Different units may rely on different information systems, vendors, intelligence feeds, or political chains of command. During a constitutional crisis, competing authorities may produce incompatible “truth” dashboards.

The struggle may therefore include disputes over:

• legal command authority;
• trusted identity credentials;
• communications access;
• intelligence databases;
• payroll and logistics systems;
• model outputs;
• emergency alert systems.

This is one reason transitions need legal continuity planning rather than improvised seizure of infrastructure.

5. The “human override” becomes constitutional, not merely technical

Many AI-governance frameworks use the phrase human-in-the-loop. In a political crisis that is insufficient. The question is which human, accountable to whom, under what law?

A meaningful democratic override requires:

• identified decision authority;
• recorded reasons;
• auditability;
• ability to refuse unlawful orders;
• independent review;
• protection for lawful whistleblowing;
• consequences for misuse.

6. What happens to security institutions after the old order falls?

Three bad models recur:

Total continuity

All personnel and command structures stay in place. Risk: the old coercive network can block reform or restore former political patrons.

Total purge

Large numbers are dismissed because of institutional affiliation. Risk: loss of competence, grievance, insecurity, and politicized replacement.

Politicized capture

The successor replaces old loyalists with new loyalists. Risk: the security sector remains partisan; only the beneficiary changes.

The democratic alternative is security sector reform (SSR) grounded in law, civilian oversight, professional standards, and individualized accountability. DCAF treats SSR as central to democratic governance after war or authoritarian rule.

7. Individualized vetting

The strongest model separates:

• responsibility for policy design;
• command responsibility;
• direct participation in serious abuse;
• concealment or falsification;
• ordinary lawful service;
• refusal or whistleblowing;
• professional competence.

Past political affiliation alone should not determine eligibility. Evidence of conduct should.

Where a person committed prosecutable crimes, criminal process is appropriate. Where conduct demonstrates serious integrity deficits incompatible with sensitive office but does not meet a criminal threshold, narrowly designed vetting may be appropriate with notice, evidence, hearing, appeal, and proportionality.

8. Intelligence agencies are the hardest institutions to democratize

DCAF describes intelligence accountability as a recurring “last frontier” of democratization. Intelligence agencies possess secrecy, privileged information, and institutional incentives to preserve capabilities.

Post-transition safeguards need:

• statutory mandates;
• independent inspectors;
• legislative intelligence committees;
• judicial authorization for intrusive powers;
• records retention and audit logs;
• prohibitions on domestic political surveillance without lawful predicate;
• protected reporting mechanisms;
• periodic declassification where possible.

9. Preventing restoration through the coercive network

An old political system can return even after electoral defeat if it retains hidden control over security institutions.

Prevention therefore requires changing chains of accountability, not merely names at the top:

• transparent appointment rules;
• professional promotion criteria;
• rotation where patronage networks are entrenched;
• clear separation of party organizations from security command;
• procurement review;
• independent complaints bodies;
• prosecution of unlawful violence;
• protection of lawful refusal.

These standards should apply regardless of whether implicated actors are associated with Democratic, Republican, military, corporate, or other networks.

10. Prediction: the strongest democratic security sector will be less opaque, not weaker

A common fear is that accountability reduces security effectiveness. The better prediction is that legitimacy increases resilience. Personnel who know the legal boundaries, record their decisions, and can refuse unlawful orders are less likely to be dragged into partisan crisis.

11. Strongest conclusion

Machine intelligence will accelerate the security state, but it does not answer the oldest political question: who has the lawful authority to command force? A durable democratic transition must preserve necessary public safety while severing partisan chains of command, auditing AI-assisted decision systems, and holding individuals accountable for conduct rather than treating entire institutions or political constituencies as collectively guilty.

Sources

1. DCAF, Post-war Security Sector Reform (2026): https://www.dcaf.ch/node/24385
2. DCAF, Intelligence Accountability: https://www.dcaf.ch/intelligence-accountability
3. DCAF, “What have we learned in the last 5 years?”: https://www.dcaf.ch/node/24186
4. OHCHR, Rule-of-Law Tools for Post-Conflict States: Vetting: https://www.ohchr.org/Documents/Publications/RuleoflawVettingen.pdf
5. Carnegie Endowment, Democratic Recovery After Significant Backsliding (2025): https://carnegieendowment.org/research/2025/04/democratic-recovery-after-significant-backsliding-emergent-lessons