Level 5 of 5Aligned to NIST CSF 2.0
Preempt
NIST CSF 2.0 Implementation Tier: Beyond Tier 4 The CSF has four tiers; this level goes past the Tier 4 description. The crosswalk
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.
-
Level 1
NIST CSF 2.0 Implementation Tier: Tier 1 · Partial
The program meets the training requirement and counts activity: training completion and simulation clicks.
-
Level 2
NIST CSF 2.0 Implementation Tier: Tier 2 · Risk Informed
The program observes behavior across several risk domains and knows who reports.
-
Level 3
NIST CSF 2.0 Implementation Tier: Tier 3 · Repeatable
Threat, access and behavior are joined per person, and the riskiest cohorts are ranked and targeted first.
-
Level 4
NIST CSF 2.0 Implementation Tier: Tier 4 · Adaptive
Playbooks orchestrate controls across the stack as risk changes, and the program proves its lift.
-
Level 5 · You are here
NIST CSF 2.0 Implementation Tier: Beyond Tier 4
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.
Level 5 · Preempt
What a level 5 program looks like.
The program acts before the attack lands. People, AI agents and vendor accounts sit on one risk model, AI predicts which sequences are forming in every priority breach scenario, and the program interrupts them before their final step, at machine speed. Access follows a subject’s stage continuously, and board key risk indicators carry thresholds tied to a stated risk appetite.
Day to day, the program sees every subject (person, AI agent, vendor account) on one record with the same threat, access and behavior fields, all nine risk domains live, and signal freshness reported as a key risk indicator. It decides from forecasts: next quarter’s stage 3+ population per breach scenario, with the forecast error recorded, and loss estimates agreed with finance. Access rises and falls with stage, sequences are interrupted before their final step, and AI agents run an approval-to-retirement lifecycle enforced by automation. It reports board KRIs inside risk appetite and human risk reduction over 6, 12 and 24 months, with the method reviewed by internal audit. A board member or committee owns the KRIs, executives carry security outcomes in their performance reviews, and the whole program is reviewed independently each year.
- Typical owner
- Board risk committee and the executive team, run by the CISO
- What great looks like
- The stage 3+ population falls each quarter for every priority breach scenario, and every board KRI sits inside its threshold.
Headline metrics to track
- 01Human risk reduction over 6, 12 and 24 months
- 02Reduction in cost of human-related breaches
- 03Board KRIs within risk appetite
- 04People and AI agents on one index
Observable criteria
The six practices at level 5.
Criteria a security leader could confirm in a meeting, with the evidence to ask for and what moves each practice up.
- 01
See
Every subject on one record, all nine domains live, gaps and freshness reported.
- People, AI agents and vendor accounts sit on one per-subject record with the same threat, access and behavior fields.
- All nine risk domains have at least one live signal, and every unmeasured signal is listed as a visibility gap with an owner and a target date.
- Signal freshness is reported as a key risk indicator against a target the program set.
- A standing process adds a source when a gap is opened and retires a source that no longer moves a decision.
Evidence to ask for
The visibility-gap list with owners and target dates · The freshness KRI on the latest board report · A per-agent record with owner, scope and last action
- 02
Understand
Forecasts with recorded error; people and agents on one model; loss and appetite per breach scenario.
- The program forecasts the stage 3+ population per priority breach scenario for the next quarter and records the forecast error.
- People and AI agents are scored on one model with the same stages and the same dimensions.
- Every priority breach scenario has a loss estimate agreed with finance and a risk appetite threshold approved by the board owner.
- Model changes are versioned, and the SOC reviews them each quarter.
Evidence to ask for
Forecast against actual for the last four quarters · The model change log · The risk appetite statement with thresholds per breach scenario
- 03
Act
Access follows stage; sequences interrupted before the final step; agent lifecycle enforced.
- Standing access rises and falls with a subject’s stage, and stage 3+ people lose standing privilege inside the window the program set.
- Sequences for every priority breach scenario are interrupted before their final step, and the count is reported.
- AI agents follow an approval-to-retirement lifecycle enforced by automation, with owner attestation on a schedule.
- Every automated action has a reversal rule and a human review threshold.
Evidence to ask for
The sequence interruption log · The agent lifecycle report (approved, attested, retired) · The access change audit for stage-driven changes
- 04
Prove
Board KRIs inside risk appetite; reduction over 6, 12 and 24 months; method audited.
- Board key risk indicators for human risk have thresholds tied to a stated risk appetite and are reviewed each quarter.
- Human risk reduction is reported over 6, 12 and 24 months with the model version and method.
- People and AI agents are reported on one index.
- Internal audit or a third party has reviewed the measurement method and data lineage.
Evidence to ask for
The board pack with KRIs and thresholds · The audit report on the method · The 24-month trend
- 05
Govern
A board owner for the KRIs; executive accountability; annual independent review.
- A board member or committee owns the human risk KRIs and reviews them each quarter.
- Executives carry security outcomes in their performance reviews.
- The model, the automation and the privacy controls are reviewed independently each year.
- Crossing a risk appetite threshold triggers a defined escalation with a named owner.
Evidence to ask for
The board agenda item and minutes · The independent review report · The escalation procedure
- 06
Culture
Culture KRIs inside appetite; executives measured; agent owners accountable.
- Security NPS, perceived friction and time to report are board KRIs with thresholds.
- Each executive sees their own MFA, reporting and simulation results each quarter.
- The gap between self-rated ability and results sits inside the target the program set.
- People who own AI agents are measured on agent hygiene (owner current, scopes reviewed, logging on) like any other behavior.
Evidence to ask for
The board pack culture KRIs · The executive behavior report · The agent-owner scorecard
Crosswalk
NIST CSF 2.0: Beyond Tier 4
SANS: Level 5 · Optimization and Resilience
The CSF has four tiers. Level 5 adds practices that go past the Tier 4 description (people and AI agents on one model, predicted sequences interrupted before incidents, KRIs inside risk appetite), so the model maps it as beyond Tier 4 and says so on the page.
The full crosswalk is in the methodology.
"Level 5: Preempt", Human Risk Management Framework v2.0.0, Living Security, CC BY 4.0. https://www.humanriskmanagement.com/maturity-model/level-5
HRM Maturity Assessment
Where does your program stand?
An initial assessment: 18 questions, about 4 minutes. Your level, six practice scores and a 90-day roadmap, shown immediately.
- Questions
- 18
- Minutes, about
- 4