About the programme

Algorithmic Immersion Radicalisation

AIR is an open research programme examining how recorded modern violence and its documented digital pathways have changed across the Five Eyes—and how those patterns sit within a wider, changing social context.

Research status
Exploratory and empirical
Geographic scope
Five Eyes
Recorded period
2000–2026 2026 is partial
Access
Free and public

A research programme, not a predetermined conclusion

The AIR Research Programme brings together a versioned public-source case database, a country-year social-context database and an explicit conceptual framework. Its purpose is to make patterns inspectable, methods contestable and uncertainty visible.

AIR asks whether documented online curation, immersion and reinforcement can help explain some pathways to coercive or violent mobilisation. It does not assume that social media exposure causes violence, that any identity group is inherently risky, or that national trends explain why a particular person acted.

Central safeguard

Contextual co-movement, chronological overlap and case-level online evidence are different forms of evidence. No single form of evidence, alone, establishes an algorithmic causal effect.

Questions the infrastructure is designed to test

01

Change over time

How have recorded terroristic attacks, thwarted plots and CIVM incidents changed by place, category and documented AIR evidence?

02

Pathways and outcomes

Which documented online behaviours, spaces and mobilisation patterns recur across otherwise different cases?

03

Social context

How have measured digital, institutional, demographic, health, economic and public-safety conditions changed within each country?

04

Alternative explanations

Which observed patterns may reflect law, disclosure, media attention, reporting practices, case selection or other non-causal data processes?

The proposed AIR cycle

The AIR cycle organises a possible pathway into four stages: curation, immersion, reinforcement and mobilisation. It is a framework for generating testable propositions, not a claim that exposure necessarily progresses to harm.

Historical-figure boundary. This early programme illustration is retained because it communicates the original AIR concept. Wording embedded in the image concerning “neurological habit reinforcement” is not an empirical measure or a finding about an individual’s neurological state. The current, testable and non-clinical operational language on the AIR Framework page governs.
Historical conceptual graphic · current operational wording governs Circular AIR model showing curation, immersion, reinforcement and mobilisation as four linked stages.
Figure 1. Original AIR-cycle illustration. A proposed cyclical pathway that may arrest, reverse or recur at any stage. It is retained as programme history; progression is not inevitable and the current operational framework supersedes embedded wording that cannot be measured by this dataset.
Read the figure description

The model begins with a proposed curation stage, followed by immersion, reinforcement and mobilisation. These are theory-led domains for investigation rather than observed states or a universal sequence. Arrows show the original cyclical hypothesis; they do not establish progression, direction, neurological change or a causal algorithmic effect.

Lawful engagement, CIVM and terrorism are not interchangeable

The programme separates protected democratic participation from coercive or violent mobilisation. CIVM is a project-defined analytical category for coercive, rights-infringing or violent ideological mobilisation that does not meet the project’s terrorism threshold. Terroristic attacks and thwarted terrorist plots remain separately coded.

Historical-figure boundary. This original illustration uses a directional layout and includes “serious organised crime” in its third tier. Neither feature governs the current case taxonomy: the tiers are non-directional conduct boundaries, and serious organised crime is not an AIR incident category. The current framework supersedes the embedded wording.
Historical conceptual graphic · non-directional boundaries govern Three tiers distinguish lawful democratic engagement, CIVM, and violent extremism or terrorism.
Figure 2. Original three-tier illustration. The retained image distinguishes lawful democratic engagement, CIVM and terrorism-related conduct. Its arrows do not imply that lawful protest develops into violence, and its embedded reference to serious organised crime is not part of the current taxonomy.
Read the figure description

Tier 1 represents lawful protest, activism and democratic engagement. Tier 2 represents qualifying CIVM conduct. Tier 3 represents terroristic attacks or thwarted terrorist plots under the current project rules. The layout is explanatory rather than developmental: it does not treat lawful participation as an early stage of violence.

Examine the full AIR Framework and evidence-level rubric

Two governed public research datasets

The Social Context v1.5 release contains public country-year observations with indicator-level definitions, units, sources, coverage and comparison restrictions. The project lead-reviewed Case Database v4.2.3 releases 346 included records through a deliberately limited 22-field public schema.

Master case records
347
authoritative workbook
Public case rows
346
22 approved fields
Populated downloadable series
78
76 explorer; 2 methodology-only
Versioned context observations
4,141
4,017 explorer; 124 methodology-only

These counts describe the research database, not population incidence. Data completeness, legal definitions, public disclosure and source availability differ across countries and time. Every public release therefore requires a version, validation record and limitations statement.

Explore case dataExplore social contextReflect on your digital useReview the methodology

A socio-technical timeline, not a causal timeline

Selected technology, platform, policy and event milestones help users orient the recorded period. The phase labels are analytical heuristics. Their placement beside case patterns does not establish that one produced the other.

Historical-figure boundary. The image is retained as an early orientation graphic. Its era names—including “Peak AIR Attack Era” and “Foreign Influence & Loneliness Era”—are programme heuristics rather than measured regimes or causal breakpoints. The interactive temporal layer uses separately dated events and optional overlays.
Historical conceptual graphic · “AIR risk” is not a validated risk score Historical conceptual timeline from 2001 to 2026 comparing selected technology milestones with AIR dataset, event and policy milestones; its AIR-risk wording is not a validated risk score.
Figure 3. Original socio-technical context timeline, 2001–2026. The image’s “AIR risk” wording is retained as programme history and must not be read as a validated risk score. Phase boundaries are approximate; platform architectures, user behaviour and incident patterns develop continuously.
Read the figure description

The timeline places selected platform and technology milestones above a 2001–2026 axis and selected AIR dataset, event and policy milestones below it. It labels an early social-media era, algorithmic-curation era, peak AIR attack era, COVID and BLM acceleration, and a foreign-influence and loneliness era. The figure notes that these are analytical heuristics rather than hard or causal boundaries.

How to interpret material on this site

Observed record

Information represented directly in a released dataset or cited public source.

Statistical description

A reproducible calculation from identified records, with its denominator and data version.

Interpretation

The programme’s reasoned reading of evidence, labelled and considered alongside credible alternatives.

Open question

A hypothesis or unresolved research problem that the current evidence does not establish.

This source-to-claim key complements the four labels used for fixed AIR Programme Insights. An Observed finding must be reproducible from observed records or statistical description; a Statistical association requires a disclosed analysis; an Informed hypothesis is an interpretation not causally established; and a Methodological or contextual observation identifies a pattern that may depend on coverage, classification, reporting or data generation. The two vocabularies describe different stages and are not interchangeable.

Religion, immigration pathway, age and other released personal descriptors are treated as descriptive case characteristics. They are not presented as causes or population-level risk factors. The public case records use a deliberately limited 22-field schema, and small sensitive aggregate cells require suppression.

Preferred website citation

Sundberg, K.W. & Reynaud, A. (2026). Algorithmic Immersion Radicalisation (AIR): Five Eyes Research Dashboard. fiveeyesair.org.

Dataset downloads carry separate release metadata and should be cited with their specific version. Individual research outputs will carry their own preferred citation only after publication.