Conceptual and operational framework

How AIR organises a proposed pathway to mobilisation

Algorithmic Immersion Radicalisation is a project-defined framework for examining documented digital pathways. It connects a proposed cycle of curation, immersion and reinforcement with mobilisation outcomes, while keeping lawful participation, CIVM and terrorism analytically distinct.

Framework status
Exploratory and testable
AIR describes
Documented pathway evidence
AIR does not describe
Culpability or severity
Temporal rule
Pre-2005 is Nil Eligibility after 2005 is not evidence by itself

A pathway lens, not a universal theory of violence

AIR proposes that digitally curated content can contribute to a cycle in which a person or bounded group encounters, repeatedly engages with and reinforces a narrowing worldview before mobilisation. The pathway may be self-directed, socially reinforced, externally encouraged or some combination of these. External orchestration is therefore possible, but it is not required for an AIR classification.

The framework is applied across terroristic attacks, thwarted terrorist plots and CIVM incidents. It asks what the reliable public record documents about online involvement. It does not assume that using the internet, holding a grievance or encountering polarising material is sufficient to explain an offence.

Non-deterministic by design

The cycle may arrest, reverse or recur. Most people exposed to contentious or extreme material do not mobilise to coercion or violence, and the current data cannot estimate an individual probability of doing so.

Curation, immersion, reinforcement and mobilisation

The four stages organise evidence and research questions. They are not a clinical sequence, a diagnostic instrument or a claim that each case passes through every stage.

  1. Proposed stage 1

    Curation

    Content is encountered through deliberate choice, social sharing and platform ranking. AIR asks what the case record documents; it does not assume that ranking narrowed exposure.

  2. Proposed stage 2

    Immersion

    Sustained engagement, repetition and digital-subculture participation may be recorded. This is behavioural evidence, not a claim about a person’s neurological state.

  3. Proposed stage 3

    Reinforcement

    Sources may document increasing certainty, identity fusion, moral disengagement or out-group hostility. The mechanism remains a hypothesis unless case evidence supports it.

  4. Proposed stage 4

    Mobilisation

    Action may range from qualifying CIVM conduct to a thwarted plot or terroristic attack. Most exposure does not lead to mobilisation, and no stage is inevitable.

Possible feedback, arrest or reversalThe four stages are analytical prompts, not a required sequence.
Figure 1. A testable representation of the proposed AIR cycle. The model distinguishes documented digital-pathway evidence from assumptions about exposure, cognition or causation.
Read the four stages in text

Curation concerns the selection and surfacing of content, including emotionally charged and identity-affirming material. Immersion concerns sustained engagement, repetition and possible digital-subculture adoption. Reinforcement concerns a possible hardening of beliefs, moral disengagement or reduced willingness to revise them. Mobilisation concerns action, which may range from coercive or rights-infringing conduct to violent extremism or terrorism. A case need not display every stage, and stages may overlap.

Protected participation is not CIVM

The three-tier framework separates lawful democratic engagement from qualifying coercive or violent conduct. It is designed to prevent the mere presence of a controversial ideology, protest movement or political identity from becoming an incident classification.

Outside the case universe

Lawful participation

Protected expression, assembly, advocacy, protest and democratic engagement remain outside the dataset, however disruptive, unpopular or morally contested.

Primary category

CIVM incident

Bounded coercive, seriously rights-infringing or violent ideological mobilisation that does not meet the project’s operational terrorism threshold.

Primary categories

Terroristic attack or thwarted plot

Executed or disrupted conduct meeting the published project test for terrorism. Legal stage and statutory treatment remain separately recorded.

Figure 2. Non-directional classification boundaries. The panels do not describe a progression from protest to violence. They separate excluded lawful conduct from the three mutually exclusive case categories.
How the tier boundary is applied

A bounded incident requires source-supported conduct. Lawful protest, advocacy and democratic participation remain outside CIVM. For protest-related cases, the dataset represents identified, arrested, charged, convicted or otherwise specifically documented participants where the sources permit; it does not classify the wider lawful crowd as offenders.

Terrorism, thwarted plots and CIVM answer different questions

Primary incident category

Terroristic attack

An executed incident coded under the project’s operational terrorism rule. The conduct and purpose must be supported by the available public record, including a function of intimidation or coercion beyond the immediate victims. Legal findings and official treatment are recorded, but the project label is not a substitute for a court determination.

Primary incident category

Thwarted terrorist plot

Preparatory or attempted conduct disrupted before the intended attack was completed. Allegations, charges, convictions and final outcomes are kept distinct, and disruption does not prove every alleged feature of a plan.

Project-defined incident category

CIVM incident

Coercive, rights-infringing or violent ideological mobilisation that does not meet the project’s terrorism threshold. The minimum screen requires qualifying conduct, an ideological or identity-related mobilisation context, a bounded actor or group and documented online activity or online-enabled group mobilisation.

Evidence dimension across categories

AIR evidence level

The project’s assessment of documented algorithmic, social-media or online involvement. AIR is not a fourth incident category. A terroristic attack, thwarted plot or CIVM incident can each have Nil, Lower, Moderate, Higher or Unknown / Not Established pathway evidence.

Primary incident categories are mutually exclusive in published totals. Separate overlap fields may record analytically relevant features across the terrorism–CIVM boundary; those fields do not change the primary counting category.

Organisation designation rule. Current listing by at least two Five Eyes governments is required before the project treats an organisation as a listed terrorist organisation for cross-jurisdictional coding. Designation is relevant but not conclusive: conduct, purpose and the case-specific public record still require assessment.

Four levels of documented AIR evidence

The levels are an unvalidated proxy for the strength and character of publicly documented digital-pathway involvement. They combine evidence that may concern exposure, communication, participation, production or offence-related dissemination. They are expert exploratory classifications, not direct measures of algorithmic exposure, psychological immersion or causal effect.

  1. Nil

    No qualifying algorithmic or social-media immersion under the project rule. This includes every pre-2005 incident, as well as source-supported offline pathways. Operational internet, email or coded communication alone does not automatically establish immersion.

  2. Lower

    Limited or incidental case-specific online involvement, such as isolated browsing, communication or contact, without evidence of sustained immersion.

  3. Moderate

    Substantive online context, including videos, propaganda, chat groups or source-supported group-level online organisation, without case-specific evidence of sustained active participation.

  4. Higher

    Case-specific sustained or active online involvement, including content production, recruitment, group administration, livestreaming or documented self-directed online immersion. A manifesto, post or livestream establishes digital activity; its timing and content must be assessed before it is interpreted as evidence of a prior immersion pathway.

Unknown / Not Established remains visible when the reviewed public record does not support a defensible level. A completed review does not turn an undisclosed pathway into a known fact.

Not a ranking of harm

Higher does not mean more violent, more culpable or more dangerous. Nil does not mean benign. The label refers only to the documented online pathway evidence under this rubric.

The conservative pre-2005 Nil rule

Every incident before 1 January 2005 is coded Nil AIR. Incidents from 2005 onwards are eligible for evidence-based Lower, Moderate or Higher coding, but the date alone supplies no evidence of online influence.

The boundary is a project rule chosen to avoid projecting the later social-media environment backwards. It does not claim that all platforms adopted personalised recommendation systems at the same moment. Platform feeds and recommendation systems developed at different times, while smartphone and short-form-video adoption followed different national and age patterns. Case-specific evidence remains necessary throughout the eligible period.

Authoritative markers used to contextualise the eras

The temporal layer uses dated milestones as contextual markers rather than causal breakpoints. Primary examples include Facebook’s 2006 News Feed documentation, Apple’s 2007 iPhone launch notice, YouTube’s 2012 recommendation-system explanation and TikTok’s 2018 global-app announcement. These sources date product changes; they do not establish effects on violence.

Four statement types keep findings and hypotheses separate

Observed finding

A directly represented value, case characteristic or reproducible descriptive result from a named release.

Statistical association

A measured relationship between variables with its sample, coverage and method disclosed. Association is not causation.

Informed hypothesis

A theory-led or evidence-informed proposition that requires further testing and could be revised or rejected.

Methodological or contextual observation

A statement about measurement, sources, legal context, reporting practice or the socio-technical environment that affects interpretation.

These labels apply to fixed AIR Programme Insights as well as narrative research outputs. Null, contradictory and complicating results belong in the record alongside supportive findings.

What the framework does not establish

  • No individual causal proof. A documented online pathway does not establish that an algorithm, platform or item of content caused the incident.
  • No demographic risk inference. Religion, migration, citizenship, sex, gender, age and other descriptors are not AIR indicators and cannot be used to infer dangerousness or population propensity.
  • No diagnosis. AIR is not a clinical assessment, mental-health classification or screening instrument.
  • No exposure threshold. Ordinary internet use, screen time or platform membership does not by itself establish Lower, Moderate or Higher AIR.
  • No inevitability. The cycle is non-linear and may arrest or reverse; lawful dissent is not treated as an early stage of violence.
  • No ecological shortcut. Country-year social-context measures describe national conditions, not what an actor saw, believed or experienced.
Read the complete methodologyExplore case dataDownload case data