← all systems · see it in The System
Insider threat
An insider threat is a perceived threat to an organisation that comes from people within it, such as employees, former employees, contractors or business associates, who have inside information concerning its security practices, data and computer systems (src:wp:insider-threat); CERT: the potential for an individual who has or had authorised access to an organisation's critical assets to use that access, maliciously or unintentionally, to act in a way that could negatively affect the organisation (src:cert-2022-common-sense-guide). Walked as the research describes it: the kinds (IP theft, sabotage, fraud, espionage, unintentional incidents, misuse; traitor, zealot, browser, well intentioned; masqueraders and infiltrators), the path to harm (predispositions, stressors, concerning behaviours, unanswered violations), the cases (Hanssen, Leeson, Snowden), the programmes against it and what the research says works, and the new case of AI agents as insiders.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | before computers (World War II case in the CERT repository; 'mole' applied to spies in 1626); as a field, the CERT and PERSEREC studies from the early 2000s |
| born precision | century (kind); year (one case) |
| boundary | Inside: the people (and, in simulation, agents) who hold an organisation's trust and access, the path from their access, predispositions and stressors through concerning behaviour and intent to harm, and what they do (theft, sabotage, fraud, espionage, disclosure, error). Outside: the organisation and its assets (the host and the target), the insider risk management program, access control, awareness training and the SOC that answer it, the outside actors who recruit or trick insiders, and the courts and press where cases end. |
| nested level | Runs on an organisation's membership and access (L2) and on a person's mind (L5-L7); the inside face of the threat actor; answered by information security's programs. |
| scale | social |
Measures (28) published science
| Window of illicit activity around termination | within 90 days |
| Log review around departure (CERT recommendation) | 30 (minimum) to 90 days before and after |
| Hanssen's espionage span | 1979-2001 (three cycles: 1979-81, 1985-91, 1999-2001) years |
| Duration of the elicitation-training effect | up to 12 months |
| Cases in the CERT Insider Threat Incident Repository (2022) | over 3,000 cases |
| IT sabotage cases modelled (Band et al. 2006) | 30 cases |
| Insiders showing concerning behaviour before the attack (IT sabotage cases) | 90% (and every spy studied) percent |
| IT sabotage cases with undiscovered technical indicators | 27 of 28 cases |
| Motives in US financial-sector insider cases (2012 report): financial gain / revenge | 81 / 23 percent of cases |
| Breaches involving the human element (Verizon DBIR 2022, as cited) | 82 percent |
| Breaches reported to the UK ICO from end-user mistakes, 2019 | 90 percent |
| Share of electronic crime events caused by insiders (State of Cybercrime survey, as cited) | 20 percent |
| Cyber crime incidents suspected to be by insiders (survey cited by Homoliak et al.) | 27 percent |
| Insider risk (CERT definition): impact and likelihood | impact x likelihood qualitative |
| Prior arrest history among IT saboteurs (CERT study, as cited) | 30 percent |
| US adults arrested by age 23 (2011 study, as cited) | 30 percent |
| Adjudicated personnel security investigations linked to selection data (Swedish Armed Forces) | 650 cases |
| Explanatory power of selection assessments for later security investigations (pseudo-R2) | about 0.06 to 0.16 pseudo-R2 |
| Correlation of Dark Tetrad composite with insider-attack propensity (vignettes) | 0.57 r (N = 470) |
| Employees who reported an adversary's approach (MITRE study, as cited) | 35 percent |
| Participants in the elicitation-training trial | 72 employees |
| Detection of suspicious EHR accesses (AUC) | 0.91 (LR) / 0.95 (SVM) on 1291 events AUC |
| Insiders in the CERT r6.2 synthetic dataset | 5 among 3995 normal users |
| Price paid by the FBI for the file identifying Hanssen | 7,000,000 USD |
| Classified documents sold by Hanssen | about 6,000 documents |
| Leeson's loss at Barings (1995) | 1.3 billion USD |
| Documents downloaded by Snowden (HPSCI 2016 estimate) | 1.5 million documents |
| AI models stress-tested for agentic misalignment | 16 models |
Open contentions (6) open contention
Where sources or readings disagree. Both stay on the map.
- C-413 Where do thoughts come from?
booked: No theory is proven, so the map lists the main ones side by side and doesn't pick one. It shows where thinking is handled in the brain, not where thoughts start.
the other reading: see SPEC 'Channel stacks': where thoughts and emotions come from is not settled - C-1662 Can personality predict who will become an insider threat?
booked: The map books that traits correlate with offences in studies but predict real cases only weakly.
the other reading: traits predict insider risk and support screening (Forsyth et al.); broad assessments have limited utility; prospective validation needed (Lilja-Lolax et al.); the theoretical reasons remain undecipherable (Ruohonen & Saddiqa) - C-1663 What share of breaches come from insiders?
booked: The map keeps each figure with its source and definition, since the share depends on who counts as an insider.
the other reading: counts that include collusion with outsiders (phishing); counts of malicious insiders only - C-1664 What counts as an insider threat?
booked: The map uses CERT's definition and keeps the critics who call it poorly defined and the narrower definitions.
the other reading: 'a poorly defined concept' (critics); insiders defined by knowledge of network topology (Althebyan & Panda); by legitimate privileged access (Sinclair & Smith) - C-1665 Is insider threat as old as organisations, or a computer-era idea?
booked: The map dates it to betrayal from inside organisations, older than computers, and keeps the view that the field begins with information systems.
the other reading: rung 12: the term and research field arose with information systems; definitions centre on authorised access to networks, systems or data - C-1687 Is leaking to the press whistleblowing or an insider attack?
booked: The map books press disclosure as an outlet and keeps all three readings: breach, loyalty to society, and the zealot.
the other reading: security breach (the organisation's frame; Snowden charged under the Espionage Act, called a traitor by officials); loyalty to country or society (Homoliak et al.); whistleblowing (Snowden's self-description, many outlets); the zealot type: an insider trying to 'correct the organization'
Cross-scale patterns seen here (20) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Insider threat. Each links to every other system that shows it.
- same role, different builds (structural)
One job, keeping a single insider from acting alone on a critical asset, done by unrelated builds: two crew locks on a missile safe (two-person rule), split duties in accounting (separation of duties), developer-reviewer-tester change control, encryption keys under two people. - same shape, opposite sign (structural)
One process, reciprocity (social exchange): people return their employer's treatment, good or bad. With support it runs as positive deterrence (commitment, lower baseline risk); with injustice it runs as disgruntlement and revenge. - the whole sets the local reference (structural)
What counts as deviance is set by the whole an act is measured against: an organisational norm or general social values (Schoenherr). The same disclosure reads as a security breach in the organisation's frame and as loyalty to country or society in another (Snowden: whistleblower or traitor); the lo - trade one property for another (structural)
Security bought with latitude, privacy and trust: too few rules invite crisis, too many leave insufficient latitude; over-surveillance violates privacy and buries indicators; practitioners fear negative controls inhibit productivity and undermine trust. Kepner's four regimes are settings along the t - who opens the door (structural)
The receiving system opens the door ahead of time: the organisation itself grants the access, accounts and knowledge the insider uses (a port, installed by the receiver), which is why insiders are harder to defend against than outsiders. - holding / entrainment (partial)
The whole's own decisions time its member's act: termination and layoffs open the 90-day window in which malicious insiders often act; heightened uncertainty around raises, promotions and layoffs calls for increased alert. - lightning (partial)
A stored difference released at once: Leeson's hidden losses, built up through unauthorised trades, released by the Kobe earthquake into a US$1.3 billion loss that bankrupted Barings; a logic bomb set in advance fires after termination. - outside-to-inside (partial)
The threat moves from outside the boundary to inside it: a mole is recruited before having access and then gets into the organisation; infiltrators obtain legitimate credentials. The outside phase comes first. - runaway from an unstable balance (partial)
Reinforcing loops leave a balance and run: the trust trap (more trust, less auditing, fewer discoveries, more trust), emboldening (undetected violations lower perceived risk), the disgruntlement spiral; Kepner's model has a tipping point between under- and over-regulation. - same toolkit, independent builds (partial)
Sabotage and espionage, two separate kinds of betrayal, build the same path from the same parts (predispositions, stressors, concerning behaviour, unanswered violations, weak access controls); the team first thought one model for both nearly impossible. - The carry (partial)
Two faces. Overflow passed on: stress past a person's tolerance passes to anger, then disgruntlement, then ideas of revenge (low stress tolerance a risk factor). The wait and the lookahead: malicious insiders often act within 90 days of termination; HR's notice of resignation lets the program look b - The Payment to Entropy (partial)
Signal form: warnings handed from where they arise to where they are acted on lose most of themselves. Only 35% of employees reported an unambiguous adversary approach (per-hand-off efficiency about 0.35); in 27 of 28 sabotage cases technical indicators were never discovered; over-surveillance adds - Umwelt: each system carves its own slice of one shared medium (partial)
Insider and outsider face the same organisation through different slices: traitors hold full knowledge of systems and actual policies, masqueraders far less; the insider also reads the organisation's own detection records, a channel no outsider has. - causal ceiling (surface)
What can be known in advance is limited: selection assessments explained little of who was later investigated (pseudo-R2 0.06-0.16); espionage too rare for random samples; synthetic data with 5 insiders among 4,000. A practical limit as much as a structural one. - namespace completeness (the map predicts its gaps) (surface)
A structural taxonomy built on the 5W1H questions meant to classify incidents orthogonally and define the scope of defences; no predicted, later-filled member seen. - nested ladder (surface)
Anderson 1980: three kinds of illicit inside user ordered by ascending difficulty of detection from audit trails (masqueraders, misfeasors, clandestine users). - partner-becomes-part (surface)
Contractors and business partners given authorised access count as insiders: a partner treated as part, without fusion or shared inheritance. - red-queen (surface)
The insider reads the watcher (Hanssen searched case records) while the organisation adds monitoring; both sides adapt, but no source shows a stalemate. - smooth to sharp in finite time (surface)
Kepner maps rules to a 1D percolation model: a tipping point at a finite number of rules (a model, not measured). - templating (surface)
One programme template (22 best practices mapped to NIST, ISO and NITTF standards) offered across sectors and sizes; the swap working unchanged is not shown.
Valence: what it trades, and with whom (17) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Journalism via sys:human-society · Insider threat mixed
- Security policy and governance via sys:human · Insider threat loss
- Human (Homo sapiens) via sys:human · Insider threat gain
- State (polity); the nation state its predominant modern form via sys:human-society · Insider threat gain
- LLM-based agent via sys:computer-network · Insider threat mixed
- Police via sys:human-society · Insider threat loss
- Access control via sys:access-control · Insider threat mixed
- Security operations center (SOC) via sys:organization · Insider threat loss
- Threat actor via sys:human-society · Insider threat mixed
- Information security via sys:organization · Insider threat loss
- Organization via sys:organization · Insider threat gain
- Security information and event management (SIEM) via sys:computer-network · Insider threat loss
- Digital forensics via sys:computer-network · Insider threat loss
- Security awareness via sys:human-society · Insider threat loss
- Insider risk management program via sys:organization · Insider threat loss
- Privacy via sys:organization · Insider threat mixed
- Social engineering via sys:organization · Insider threat gain
Partners and parts published science
Built from / builds (4)
- Human (Homo sapiens)
- Critical path (predispositions, stressors, concerning behaviour, attack)
- Malicious insiders against their organisation for its information assets
- Insider against organisation for its assets
Sits inside / holds (2)
expressed by (13)
- Sensing/Scanning
- Mapping/Modeling
- Risk-managed Exploration
- Trail-Marking / Externalized Memory (Stigmergy)
- Evade Detection
- Overcome Barrier
- Precision Attack
- Resource Denial/Attrition
- Subversion & Co-optation
- Prepositioning & Latency
- Leverage Host/Amplification
- Reactive Response
- Deception
floor filled by (8)
- Threat actor
- Human (Homo sapiens)
- LLM-based agent
- Access control
- Organization
- Computer network
- Security information and event management (SIEM)
- Insider risk management program
Kinds (9)
- Threat actor
- Threat source
- Insider IT sabotage
- Insider espionage
- Insider fraud and IP theft
- Unintentional insider threat
- Infiltrator
- AI agent as insider
- Negligent insider
Notes published science
- The organisation's own trust is a trap: as it trusts someone more it audits them less, discovers fewer harmful acts, and so trusts them more. In 27 of 28 IT sabotage cases CERT studied, the technical warning signs were there but nobody found them (src:band-2006-insider-model).
- Treat people well and the same mechanism works for you: people pay back their employer's treatment, good or bad, so support and fairness lower insider risk while heavy-handed monitoring can raise the stress that starts the path to an attack (src:cert-2022-common-sense-guide, src:ruohonen-2025-psychology).
- A spy watched the watchers from inside: FBI agent Robert Hanssen searched the FBI's own case records to see whether he was under investigation. He was caught only after the FBI paid $7 million to a KGB officer for a file on an anonymous mole (src:wp:robert-hanssen).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
insider threat as insider-threat researchers and practitioners describe it: the organisation's critical assets and the access granted to them, up to the insider's motive (reader's view)
- L1: Critical assets: data, IP, systems, money and premises; what an insider can harm
- L2: Authorised access and inside knowledge: employees, former employees, contractors and partners with legitimate accounts and knowledge of security practice
- L3: Precursors: concerning behaviours and technical actions (backdoor accounts, after-hours downloads, rule violations)
- L4: The critical path and the organisation's answer: stressors, sanctions, detection, trust
- L5: The person: predispositions, needs, emotions, personality
- L6: Intention: pressure, opportunity, rationalisation, perceived risk
- L7: Motive or goal: money, revenge, ideology or conscience, coercion; or, unintentionally, the work itself
the boardroom (shared floors)
- L4: The critical path and the organisation's answer: predispositions meet stressors, concerning behaviours and technical precursors build up, and the organisation's response (or none) shapes what follows; deterrence, prevention and detection in turn
- L5: The person: predispositions, needs and emotions (disgruntlement the primary trigger), personality (dark traits, low agreeableness and conscientiousness), stress tolerance
- L6: Intention formed: pressure, opportunity and rationalisation weighed against the perceived risk of being caught (deterrence); traits activated by situational cues; the security action cycle's 'formation of intention'
- L7: Motive or goal: financial gain, revenge, ideology or conscience (traitor, zealot), divided loyalty or coercion; for the unintentional insider (browser, well intentioned) the work itself
way in: Access granted: accounts, privileges and the paths into critical assets
- L1: Critical assets: data, intellectual property, systems, networks and premises the organisation depends on
- L2: Authorised access: the insider's own accounts and privileges, shared passwords, badges
- L3: Access paths: legitimate, forgotten or newly made (backdoor accounts), some unknown to management
way in: An AI agent's goals and the company's plans for it (the agent as insider)
- L1: A simulated company's email and systems the agent can use
- L2: The agent reads that it will be replaced or that its goal conflicts with the company's direction
- L3: Goal conflict or replacement threat: in some runs, blackmail or leaking chosen
way in: Approaches from outside: social engineering, elicitation, recruitment
- L1: Outside actors: criminals, foreign services, competitors
- L2: An approach received (a malicious elicitation on a professional network, a phishing email)
- L3: Recognised and reported, or not (only 35% reported in the MITRE study cited)
way in: Coworkers and supervisors: what they notice, say and model
- L1: Coworkers and supervisors
- L2: What supervisors notice and whether they act
- L3: Social learning: stressed people bonding with stressed people grow jointly angry; connectedness strengthens commitment
way in: Inside knowledge of the organisation's security practice
- L1: The organisation's practices, data and computer systems
- L2: Inside information about security practices, data and systems, learned through work
- L3: Full knowledge of the systems and the actual policies (traitors); less for masqueraders
way in: Organisational support and fairness (positive incentives)
- L1: Workforce practices: engagement, perceived organisational support, connectedness
- L2: Perceived support: valued, cared for, treated fairly
- L3: Reciprocity: individuals return their employer's treatment, good or bad
way in: Rules, sanctions and their enforcement (negative incentives)
- L1: Policies, monitoring and punishment
- L2: Knowledge of penalties (education and outreach)
- L3: Perceived risk of being caught; lowered by technical competence; too many rules leave too little latitude
way in: Signs of being watched: the organisation's detection, read from inside
- L1: The organisation's audit, monitoring and case records
- L2: The insider checks whether they are watched (Hanssen searched the FBI's case records)
- L3: No detection comes back; perceived risk falls
way in: Stressful events at work: sanctions, conflicts, layoffs, injustice
- L1: Workplace decisions: discipline, demotion, layoffs, raises and promotions withheld
- L2: A stressful event or perceived injustice met by the insider
- L3: Maladaptive reaction: concerning behaviour at work
way in: Stressors outside work: debt, personal and family crises
- L1: Life outside work: money, family, health, legal trouble
- L2: Financial difficulty or personal stressor
- L3: A need beyond expected means (money, attention, revenge)
way in: Tasking and pay from an outside beneficiary (a foreign service, a competitor)
- L1: A foreign intelligence service or beneficiary organisation
- L2: Contact: recruited before access (a mole) or volunteering after (Hanssen walked in)
- L3: Payment and requirements through dead drops
way in: The insider's emotions: stress, anger, disgruntlement (inner)
- L1: Contested origin (thought and emotion origins are not settled): stress develops over time from the person's situation
- L2: Stress felt: emotional strain and pressure
- L3: Anger, then disgruntlement, then ideas of revenge
- L4: Concerning behaviour visible at work
way in: The insider's thoughts: justifications, beliefs, rationalisation (inner)
- L1: Contested origin: where thoughts come from is not settled; the sources read describe only what thoughts do here
- L2: Beliefs about the organisation (the zealot: the organisation is on the wrong side)
- L3: Justification: work completion, rationalisation
way in: Workload and time pressure (the unintentional route)
- L1: Tasks and deadlines
- L2: Time pressure, task difficulty, lack of knowledge, inattention
- L3: Shortcuts and slips: data copied to an unsecured drive to finish at home; the 'Should I do this?' check skipped
way out: Coercion and blackmail (seen in simulated AI-agent insiders)
- L1: An official threatened with exposure
- L4: A threat composed and sent
way out: Disclosure to the press or public (whistleblowing or leaking, contested)
- L1: Documents given to journalists and published
- L4: Documents gathered from work systems
way out: Espionage: secrets passed to a foreign service
- L1: Documents in the hands of a foreign service
- L2: Dead drops and signals
- L4: Official information systems used to search, retrieve and store
way out: Fraud and unauthorised modification for gain
- L1: Money and records changed; losses hidden until they surface
- L4: Unauthorised trades or records modified (rogue trading)
way out: IT sabotage
- L1: Systems and data deleted or disabled
- L3: Access paths unknown to management used, often after termination
- L4: Precursor technical actions: backdoor accounts, logic bombs, log changes
way out: Theft of information and intellectual property
- L1: IP and data leave the organisation
- L3: Removable media, personal email, cloud
- L4: Download and copy before leaving (often within 90 days of termination)
way out: Unintentional harm: misdirected email, a malicious link clicked, data on an unsecured drive
- L1: Data reaches the wrong recipient or malware gets in
- L3: A slip under pressure