A cascaded probabilistic framework for stealth-aware MITRE ATT&CK path prediction in active directory adversary emulation

Abstract Proactive adversary emulation requires anticipating multi-step attack trajectories, yet existing approaches rely on static attack graphs or single-signal models that cannot jointly satisfy coherence, tactical realism, and evasiveness. We propose a cascaded probabilistic framework for MITRE ATT&CK technique-path prediction in Active Directory environments, where exhaustive enumeration is intractable. Its contribution is a three-stage architecture that decouples these objectives instead of blending them. Stage 1, a two-layer unidirectional LSTM trained on observed threat-actor chains, emits a five-candidate pool rather than one technique. Stage 2, an empirical Markov tactic-transition gate, admits only attested transitions, including self-loops and backward moves a strict linear kill chain excludes. Stage 3 re-ranks survivors against a detection-rule corpus, favoring evasive over merely probable techniques. Candidates are masked at inference to Windows-applicable techniques, so outputs are technique chains, not instantiated paths. Under a prefix-disjoint protocol in which no evaluation context occurs in training, the true continuation is retained in the five-candidate pool more often than by a first-order predictor or an exact prefix-lookup table on identical pairs. Because Stage 3 both selects on and is scored by detection-rule density, the chains are re-scored against independently maintained corpora the selector never accesses; the exposure reduction persists there, and it is that transferred result that carries the evasiveness claim. Novel here is that structural admissibility is a hard constraint rather than a soft weight, so tactical realism cannot be traded away for likelihood or stealth. A case study initiated from a single initial-access technique further indicates that all tactic transitions within the returned path portfolio have attested support in the corpus transition matrix, and that no predicted step violates the ATT&CK-declared phase set of its technique.

Authors

Publication Details

Journal
Scientific Reports
Published
2026-10-13
DOI
https://doi.org/10.1038/s41598-026-71934-z
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

A cascaded probabilistic framework for stealth-aware MITRE ATT&CK path prediction in active directory adversary emulation

Alshaimaa Abo‐alian, Nagwa L. Badr, Mahmoud Youssef
Scientific Reports
Network Security and Intrusion Detection
article

A cascaded probabilistic framework for stealth-aware MITRE ATT&CK path prediction in active directory adversary emulation

Alshaimaa Abo‐alian, Nagwa L. Badr, Mahmoud Youssef
article en

Abstract

Abstract Proactive adversary emulation requires anticipating multi-step attack trajectories, yet existing approaches rely on static attack graphs or single-signal models that cannot jointly satisfy coherence, tactical realism, and evasiveness. We propose a cascaded probabilistic framework for MITRE ATT&CK technique-path prediction in Active Directory environments, where exhaustive enumeration is intractable. Its contribution is a three-stage architecture that decouples these objectives instead of blending them. Stage 1, a two-layer unidirectional LSTM trained on observed threat-actor chains, emits a five-candidate pool rather than one technique. Stage 2, an empirical Markov tactic-transition gate, admits only attested transitions, including self-loops and backward moves a strict linear kill chain excludes. Stage 3 re-ranks survivors against a detection-rule corpus, favoring evasive over merely probable techniques. Candidates are masked at inference to Windows-applicable techniques, so outputs are technique chains, not instantiated paths. Under a prefix-disjoint protocol in which no evaluation context occurs in training, the true continuation is retained in the five-candidate pool more often than by a first-order predictor or an exact prefix-lookup table on identical pairs. Because Stage 3 both selects on and is scored by detection-rule density, the chains are re-scored against independently maintained corpora the selector never accesses; the exposure reduction persists there, and it is that transferred result that carries the evasiveness claim. Novel here is that structural admissibility is a hard constraint rather than a soft weight, so tactical realism cannot be traded away for likelihood or stealth. A case study initiated from a single initial-access technique further indicates that all tactic transitions within the returned path portfolio have attested support in the corpus transition matrix, and that no predicted step violates the ATT&CK-declared phase set of its technique.

Scientific ReportsVol. 16(1)
Openalex Percentile: Top 11%
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

A cascaded probabilistic framework for stealth-aware MITRE ATT&CK path prediction in active directory adversary emulation — Alshaimaa Abo‐alian, Nagwa L. Badr, et al. · Scientific Reports (2026) | TGRS Research Map | TGRS