HB-09Level 4 · Breach scenarioDBIR 2026 AI findings · NIST AI RMF
Shadow AI & agent misuse
- Level 4 · Breach scenario
- 18 signals
- 4 risk domains
A loss in which sensitive data leaves through an unsanctioned AI tool or a personal AI account, or an AI agent acts outside its scope, without logging or under a compromised owner.
Classification
- Fed by risk domains One of 9 areas of human risk, from social engineering to readiness and response. Every signal belongs to one.
- Dimensions Every signal answers one of 3 questions.
- Evidence level
- Anchor
-
- DBIR 2026 AI findings · NIST AI RMF
- NIST CSF 2.0
-
- ID.AM-02
- PR.DS-10
- GV.PO-01
- MITRE ATT&CK
-
- T1567
- T1528
-
1
Sequence
End in this breach scenario.
-
3
Cohorts
Predict it.
-
18
Signals
Feed it.
-
4
Risk domains
Where those signals come from.
Sequences that lead here
One sequence ends here.
Each one is a set of signals in order inside a window. Watch it play.
SQ-DL-023 stepsPeak: Stage 3 Exploited
Shadow AI drift
Step 01ATT&CK tactic: Collection
Unauthorized AI tool
AU-B-01Uses unauthorized AI tools
Stage 2Susceptible
Step 02ATT&CK tactic: Collection
Personal account login
AU-B-03Uses personal accounts on AI platforms
Stage 2Susceptible
Step 03ATT&CK tactic: Exfiltration
Sensitive data in prompt
AU-B-02Puts sensitive data in AI prompts
Stage 3Exploited
Shadow AI drift, 3 steps inside 30 days: Unauthorized AI tool (AU-B-01 Uses unauthorized AI tools, Collection, Stage 2 Susceptible); then Personal account login (AU-B-03 Uses personal accounts on AI platforms, Collection, Stage 2 Susceptible); then Sensitive data in prompt (AU-B-02 Puts sensitive data in AI prompts, Exfiltration, Stage 3 Exploited). It leads to HB-09 Shadow AI & agent misuse.
-
SQ-DL-02
Shadow AI drift
- Step 1: Unauthorized AI tool (AU-B-01 Uses unauthorized AI tools)
- Step 2: Personal account login (AU-B-03 Uses personal accounts on AI platforms)
- Step 3: Sensitive data in prompt (AU-B-02 Puts sensitive data in AI prompts)
- Breach scenario: HB-09 Shadow AI & agent misuse
The definition, in detail
Definition
It is anchored to the AI findings in the 2026 Verizon DBIR and to the NIST AI Risk Management Framework.
Example
A reporting agent with tenant-wide read access and no logging is driven by a prompt injection to post the finance share to a public channel.
How it is measured
- Anchor
- DBIR 2026 AI findings · NIST AI RMF
- SOC signals
- AI data loss, unsanctioned apps
- Fed by
- 18 signals across 4 risk domains
How it is used
Programs measure toward Shadow AI & agent misuse by counting the 18 signals, 3 cohorts and 1 sequence that point to it, and tie them to the incidents the SOC records as ai data loss, unsanctioned apps.
A breach scenario names what the evidence beneath it points toward. The organization’s own incidents confirm it; the signals, cohorts and sequences show where it is likely.
Technical mappings
Framework mappings; validate for your use case.
Continue exploring
"Shadow AI & agent misuse" (HB-09), Human Risk Management Framework v2.0.0, Living Security, CC BY 4.0. https://www.humanriskmanagement.com/taxonomy/breach-scenarios/hb-09-shadow-ai-and-agent-misuse
Part of the open HRM Taxonomy, CC BY 4.0. Download the taxonomy as JSON · What's new in 2.0. This page describes one breach scenario.
The initial assessment
Find your level.Then fix what matters.
18 questions, about 4 minutes and nothing to type before the result: your level, a score for each practice and three first playbooks, each Easy, Medium or Hard.