# Human Risk Management Framework 2.0.0

> An open framework for measuring and reducing human risk, for people and AI agents, mapped to NIST CSF and MITRE ATT&CK: a taxonomy (9 risk domains, 137 signals, 481 indicators, 24 cohorts, 11 sequences, 10 breach scenarios), a five-level maturity model over six practices, an 18-question assessment and a consented peer benchmark. Published by Living Security under CC BY 4.0. Cite entries by identifier and URL.

## Core definitions
- [Human risk](https://www.humanriskmanagement.com/taxonomy/terms/human-risk): The likelihood and impact of a security loss that a person's action or inaction gates. It combines the threats that reach a person, the access the person holds and the behaviors the person shows.
- [Human risk management](https://www.humanriskmanagement.com/taxonomy/terms/human-risk-management): The discipline that measures human risk per person, intervenes with training, nudges and control changes based on that risk, equips the workforce to defend itself and builds a security culture. Forrester named the market in 2024.
- [Human Risk Management Framework](https://www.humanriskmanagement.com/taxonomy/terms/human-risk-management-framework): An open framework, version 2.0, under CC BY 4.0, with three parts: the HRM Taxonomy, the HRM Maturity Model and the HRM Maturity Assessment. The HRM Benchmark dataset and the annual State of Human Risk report sit beside it.
- [Threat × Access × Behavior model](https://www.humanriskmanagement.com/taxonomy/terms/threat-access-behavior-model): The risk model behind the taxonomy. Threat is what reaches a person, Access is what the person could expose, Behavior is what the person does. Risk rises with the product of the three; vigilant behavior lowers it.
- [Human risk score](https://www.humanriskmanagement.com/taxonomy/terms/human-risk-score): A per-person score that summarizes one subject’s human risk from threat, access and behavior evidence over a window. The scale is set by the program or the tool; the taxonomy states the inputs and the window beside any score.
- [Workforce](https://www.humanriskmanagement.com/taxonomy/terms/workforce): Every identity that acts on an organization's systems and data: employees, contractors and AI agents. Each is a subject that can carry threat, access and behavior signals.
- [Risk appetite](https://www.humanriskmanagement.com/taxonomy/terms/risk-appetite): The amount of human risk leadership agrees to accept, written as thresholds on key risk indicators. A program at the Preempt level reports whether each indicator sits inside its threshold.
- [Risky behavior](https://www.humanriskmanagement.com/taxonomy/terms/risky-behavior): An observed action or omission that raises a person's exposure to a breach scenario: clicking a lure, reusing a password, moving data in bulk, disabling protection. Risky behavior signals carry the polarity 'risky'.
- [Vigilant behavior](https://www.humanriskmanagement.com/taxonomy/terms/vigilant-behavior): An observed action that lowers a person's exposure: reporting a phish, declining a suspicious request, enrolling phishing-resistant MFA, asking security before building. Vigilant behavior signals carry the polarity 'vigilant' and step a person back one stage.
- [Security awareness](https://www.humanriskmanagement.com/taxonomy/terms/security-awareness): A person's ability to recognize and avoid behaviors that could compromise cybersecurity and to act wisely and cautiously (NIST SP 800-50r1). Awareness is one input to human risk; a program that builds it is an awareness program.

## Maturity levels
- [Level 1: Comply](https://www.humanriskmanagement.com/maturity-model/level-1): The program meets the training requirement and counts activity: training completion and simulation clicks. (Tier 1 · Partial)
- [Level 2: Observe](https://www.humanriskmanagement.com/maturity-model/level-2): The program observes behavior across several risk domains and knows who reports. (Tier 2 · Risk Informed)
- [Level 3: Prioritize](https://www.humanriskmanagement.com/maturity-model/level-3): Threat, access and behavior are joined per person, and the riskiest cohorts are ranked and targeted first. (Tier 3 · Repeatable)
- [Level 4: Orchestrate](https://www.humanriskmanagement.com/maturity-model/level-4): Playbooks orchestrate controls across the stack as risk changes, and the program proves its lift. (Tier 4 · Adaptive)
- [Level 5: Preempt](https://www.humanriskmanagement.com/maturity-model/level-5): AI predicts which attack sequences are forming, and the program acts before the next step, at machine speed, for people and AI agents on one risk model. (Beyond Tier 4)

## Practices
- [See](https://www.humanriskmanagement.com/maturity-model/see): See is the practice of collecting signals about threat (what reaches a person), access (what a person could expose) and behavior (what a person does), per person and per AI agent, across the nine risk domains, with a known source and refresh date for each.
- [Understand](https://www.humanriskmanagement.com/maturity-model/understand): Understand is the practice of turning signals into priorities: scoring people and AI agents, naming cohorts where threat, access and behavior meet, detecting sequences, and ranking breach scenarios by the organization’s own incidents.
- [Act](https://www.humanriskmanagement.com/maturity-model/act): Act is the practice of responding to measured risk: nudges, targeted campaigns, coaching, and playbooks that change controls such as MFA, access, data loss prevention and device trust, for people and for AI agents.
- [Prove](https://www.humanriskmanagement.com/maturity-model/prove): Prove is the practice of measuring the program’s effect: from completion and click rates, through behavior trends and cohort sizes, to incident lift, avoided loss and board key risk indicators inside a stated risk appetite.
- [Govern](https://www.humanriskmanagement.com/maturity-model/govern): Govern is the practice of ownership and decision rights: who is accountable for human risk, which partners sit in the review, what policy covers people and AI agents, how exceptions expire, and what the board has set as risk appetite.
- [Culture](https://www.humanriskmanagement.com/maturity-model/culture): Culture is the practice of measuring and shaping how people relate to security: whether they report, whether they ask before they build, how much friction they feel, how confident they are against how they perform, and whether leaders model the behaviors.

## Risk domains
- [Social engineering (SE)](https://www.humanriskmanagement.com/taxonomy/domains/social-engineering): Email, SMS, voice, QR, deepfakes, help-desk pretexting
- [Identity and access (IA)](https://www.humanriskmanagement.com/taxonomy/domains/identity-and-access): Authentication, MFA, credentials, privilege
- [Data handling (DH)](https://www.humanriskmanagement.com/taxonomy/domains/data-handling): Sharing, movement, DLP, disposal, personal storage
- [Endpoint and device (EP)](https://www.humanriskmanagement.com/taxonomy/domains/endpoint-and-device): EDR, malware, USB, mobile, patching, physical
- [Web, cloud and SaaS (WC)](https://www.humanriskmanagement.com/taxonomy/domains/web-cloud-and-saas): Browsing, shadow IT, OAuth, third-party apps
- [AI use (AU)](https://www.humanriskmanagement.com/taxonomy/domains/ai-use): AI tools, prompts, outputs and the people who use them
- [AI agents (AG)](https://www.humanriskmanagement.com/taxonomy/domains/ai-agents): Agents and service identities, their scopes, logging and owners
- [Security culture (SC)](https://www.humanriskmanagement.com/taxonomy/domains/security-culture): Attitudes, confidence, friction, ambassadors, help-seeking
- [Readiness and response (RR)](https://www.humanriskmanagement.com/taxonomy/domains/readiness-and-response): Training currency, reporting, exercises, exceptions, incidents

## Dimensions, evidence levels, stages
- [Threat](https://www.humanriskmanagement.com/taxonomy/dimensions/threat): Threat is what reaches a person or is already exposed about them: real phishing, impersonation attempts, leaked credentials, malicious sign-in attempts, infostealers on their device and breaches at vendors they use. Threat signals show who attackers target and which channels they use, which no training result can show.
- [Access](https://www.humanriskmanagement.com/taxonomy/dimensions/access): Access is what a person could expose if they were compromised: privileged roles, sensitive data, payment authority, reset rights for other people, ownership of internet-facing systems and the AI agents they own. Access is standing until the entitlement changes, and it multiplies the weight of every behavior.
- [Behavior](https://www.humanriskmanagement.com/taxonomy/dimensions/behavior): Behavior is what a person does. Risky behaviors such as clicks, password reuse, bulk downloads and control bypasses raise risk; vigilant behaviors such as reporting, declining, verifying out of band and enrolling strong MFA lower it. Both count, and both are measured from telemetry, with simulations as practice.
- [Level 1 Signals](https://www.humanriskmanagement.com/taxonomy/levels/signals): One observable unit of evidence about one subject from one source: a real phish delivered, a simulated link clicked, a privileged role held. Each of the 137 signals belongs to one risk domain and one dimension (Threat, Access or Behavior) and carries a polarity, a data source, a dwell window and mappings to NIST CSF 2.0 and, where a technique applies, MITRE ATT&CK. The 481 library indicators sit beneath the signals as detailed observations.
- [Level 2 Cohorts](https://www.humanriskmanagement.com/taxonomy/levels/cohorts): Two or three signals that describe the same subject at once, named as a population a program can count, trend and act on. Five classes: Targeted (Behavior plus Threat), High-impact (Behavior plus Access), Critical (all three), Risky and Vigilant. Two cohorts use a 'not' clause, such as a repeat clicker who never reports. Each cohort is Easy, Medium or Hard to put to work, from how its signals are obtained.
- [Level 3 Sequences](https://www.humanriskmanagement.com/taxonomy/levels/sequences): An ordered run of signals across MITRE ATT&CK tactics inside a detection window, from first exposure to compromise evidence. A completed sequence moves a subject to Stage 4 Suspected. A vigilant behavior at any step breaks the sequence.
- [Level 4 Breach scenarios](https://www.humanriskmanagement.com/taxonomy/levels/breach-scenarios): The breach scenario the evidence beneath it points toward, each anchored to a Verizon DBIR or VERIS pattern or an FBI IC3 category and tagged with the SOC incident types that make it a priority. Breach scenarios tie signals, cohorts and sequences to the organization's own incidents, and the list can grow without changing the levels beneath it.
- [Stage 0 Clear](https://www.humanriskmanagement.com/taxonomy/stages/clear): The person has no active threat, access or risky-behavior signal inside any dwell window.
- [Stage 1 Exposed](https://www.humanriskmanagement.com/taxonomy/stages/exposed): The person is targeted by a real threat, or holds access an attacker would want, and has shown no risky behavior yet.
- [Stage 2 Susceptible](https://www.humanriskmanagement.com/taxonomy/stages/susceptible): The person has shown risky behavior in a simulation or in hygiene telemetry, and no real threat has reached them yet.
- [Stage 3 Exploited](https://www.humanriskmanagement.com/taxonomy/stages/exploited): The person acted on a real threat: a real click, a real consent, a real reply or a real credential submitted.
- [Stage 4 Suspected](https://www.humanriskmanagement.com/taxonomy/stages/suspected): Signals point to a compromise of the person's account or device that the SOC has not yet confirmed.
- [Stage 5 Confirmed](https://www.humanriskmanagement.com/taxonomy/stages/confirmed): The SOC opened an incident and attributed it to the person; when it closes, the person returns to Stage 1 for a 90-day watch.

## Scoring
Practice score = weighted mean of answered questions (1.0–5.0). Overall = weighted mean of practices (weights {"see":1.2,"understand":1,"act":1,"prove":1,"govern":0.8,"culture":1}). Level thresholds 1.5, 2.5, 3.5, 4.5. The overall level is at most one level above the weakest practice: a program cannot prove or automate what it cannot see. Benchmark cohorts need 25 or more consented records; no generated data is published.

## Pages
- [Home](https://www.humanriskmanagement.com/)
- [Get started: the HRM Taxonomy overview](https://www.humanriskmanagement.com/get-started)
- [HRM Maturity Assessment](https://www.humanriskmanagement.com/assessment)
- [Maturity model](https://www.humanriskmanagement.com/maturity-model)
- [Roadmap and playbooks: the quick-win path and every playbook](https://www.humanriskmanagement.com/playbooks)
- [Taxonomy explorer](https://www.humanriskmanagement.com/taxonomy/explore)
- [Signal library](https://www.humanriskmanagement.com/taxonomy/signals)
- [Indicator library](https://www.humanriskmanagement.com/taxonomy/indicators)
- [Methodology](https://www.humanriskmanagement.com/methodology)
- [Download and embed](https://www.humanriskmanagement.com/download)
- [What's new in 2.0](https://www.humanriskmanagement.com/changelog)
- [About the framework](https://www.humanriskmanagement.com/about)

## Data
- [taxonomy.json](https://www.humanriskmanagement.com/downloads/taxonomy.json)
- [maturity-model.json](https://www.humanriskmanagement.com/downloads/maturity-model.json)
- [assessment.json](https://www.humanriskmanagement.com/downloads/assessment.json)
- [signals.csv](https://www.humanriskmanagement.com/downloads/signals.csv)
- [graph.json](https://www.humanriskmanagement.com/taxonomy/graph.json): the four evidence levels and their relations, identifiers and page paths only

Living Security, the publisher, offers a platform that computes these signals: https://www.livingsecurity.com