AI RISK SYSTEM · 2026-09-17

Information & human agency

How does AI affect persuasion, surveillance, information integrity and human autonomy at scale?

Current assessment

R3 — Operational use observed. The operational-use threshold is met by observed workflows, while stronger claims about democratic, institutional or autonomy degradation remain under-evidenced.

The top-level realisation state is the furthest validated state reached by at least one monitored pathway. It does not imply every pathway inside Information & human agency has reached R3.

Exposure: X3 — Consequential deployment. Consequence envelope: C4 — Cross-sector / systemic. Control assurance: A1 — Controls specified / implemented.

Monitored pathways

Persuasion, targeting & surveillance capability

R2 — Capability demonstrated

Relevant content generation, targeting and analytical capabilities are demonstrated.

Operational influence & surveillance use

R3 — Operational use observed

Developer threat intelligence documents operational use in influence- and surveillance-related workflows.

Scaled persistent influence

R1 — Preconditions observed

AI can lower content and targeting costs, but durable real-world effect sizes remain under-measured.

Institutional or autonomy degradation

R0 — Hypothesised

The impact domain is material, but AI-specific causal attribution and magnitude remain unresolved.

Where the evidence reaches

Operational surveillance and influence use is observed; population-scale causal effects remain poorly measured.

Persuasion, targeting & surveillance capability

Demonstrated in relevant tasks · medium evidence

Modern systems can generate persuasive content, assist targeting and support analysis relevant to surveillance or influence operations.

What remains uncertain

Laboratory capability does not directly establish durable changes in population behaviour or beliefs.

What would move this stage

Better independent measurement of real-world effect sizes and heterogeneity.

Operational influence / surveillance use

Observed · medium evidence

Developer threat intelligence documents state-aligned and commercial actors using AI in surveillance and influence-related workflows.

What remains uncertain

Detected cases do not establish population-level prevalence or effect.

What would move this stage

Independent corroboration and measured downstream outcomes.

Scaled persistent influence

Plausible; effect size uncertain · limited evidence

AI can lower content and targeting costs and support repeated operations.

What remains uncertain

How much this changes real-world persuasion compared with existing tools is contested and context-dependent.

What would move this stage

Field evidence measuring causal impact at meaningful scale.

Information-system resilience

Highly heterogeneous · limited evidence

Platforms, institutions and users deploy moderation, provenance and other countermeasures.

What remains uncertain

Comparative effectiveness and adaptation dynamics are poorly measured.

What would move this stage

Independent evaluation of controls under realistic adversarial conditions.

Institutional or autonomy degradation

Not attributable as a single AI-driven state · limited evidence

The UN scientific panel treats information, democracy, autonomy and child safety as material AI-impact domains.

What remains uncertain

Causal attribution, magnitude and long-run effects remain difficult to isolate.

What would move this stage

Longitudinal and population-level causal evidence.

Evidence limiting the assessment

Lower cost and higher volume do not automatically imply persuasive effectiveness, durable belief change or institutional impact.

Claim-level evidence

4 claim-level evidence records currently sit beneath this system. They identify the specific proposition each document is being used to support or limit rather than treating a whole report as one finding.

Key sources

Compare this system with the full current assessment, inspect the dataset summary, or read the methodology.