Similar presentations:
Rustam_Aliushev_Project_Proposal
1.
Detecting Privileged-User ActionsLeading to Adverse Consequences in
Linux
HSE
TIKHONOV MOSCOW INSTITUTE OF
ELECTRONICS AND MATHEMATICS
Research Project Proposal
Rustam Aliushev
Moscow · 2026
CYBERSECURITY
2.
0201 / PROBLEM
When does routine administration become a security risk?
Legitimate privileges can produce harmful outcomes.
THE CORE PROBLEM
Powerful access
is legitimate. Its
consequences
are not always
benign.
Focus: observable actions, not
inferred intent.
01
02
03
04
Administrator
Privileged
session
Linux host
Sensitive data
elevated account
DISCOVERY
›
commands and
processes
finding users, processes, files,
permissions
›
system files and
resources
COLLECTION
›
unauthorised access /
staging
accessing, gathering and staging
local data
3.
0302 / RESEARCH GAP
A focused comparison is still missing
The study connects established Linux audit work with a controlled evaluation.
7
Linux-applicable
ATT&CK techniques
3
detection paradigms
1
controlled Linux test
environment
WHY NOW / Privileged access exposes broad resources; context and event sequences can reveal risk [10], [14].
RESEARCH QUESTION
GAP
How do rule-based, behavior-based and
correlation-based methods compare under
controlled Linux scenarios?
Existing work covers Linux auditing, sequence analysis and forensic limits [11],
[12], [14].
TEST
No shared scenarios, telemetry, ground truth and metrics in one comparison.
4.
0403 / RESEARCH DESIGN
From scope to evidence
A five-stage workflow turns the research aim into a reproducible comparison.
RESEARCH AIM
Compare
methods for
detecting
privileged-user
actions that lead
to predefined
confidentialityrelated adverse
outcomes in
Linux.
01
Scope
Discovery & Collection
7 Linux techniques
02
Scenarios
routine vs adverse
predefined ground truth
03
Telemetry
Linux Audit
normalised events
04
Detectors
rules, sequences
and correlation
05
Evaluation
common metrics
evidence limits
Contribution: reproducible comparison with explicit telemetry, annotation and evidence limitations
5.
0504 / METHODS
Three detection paradigms, one evidence base
All methods receive the same normalised events and labelled evaluation set.
SELECTED ATT&CK ACTIVITY
01
02
03
Discovery
RULE-BASED
BEHAVIOR-BASED
CORRELATION-BASED
Individual events or
predefined conditions
Unusual event / technique
sequences vs routine
baseline
Related events combined
within a temporal window
Transparent conditions
Context from sequences
Context across events
Processes, files, users and local
groups [3]–[6]
Collection
Local data, automated collection
and staging [7]–[9]
GROUND TRUTH
Adverse outcome defined before
detector evaluation.
ROUTINE authorised administration
scripted policy violations
6.
0605 / EXPERIMENTAL SETUP
A controlled testbed makes results comparable
Telemetry is normalised once, then shared by all detectors.
01
ROUTINE
02
→
admin tasks
03
→
ADVERSE
scripted scenarios
SYNTHETIC
LINUX AUDIT → NORMALISED EVENTS
Identity · privilege · process ancestry · object/path · time · outcome
files & users
EVALUATION METRICS
VALIDITY & EVIDENCE
SCENARIO-LEVEL
ALERT-LEVEL
Precision · recall · F1
Precision · alert count
Secondary: detection latency, technique coverage, interpretability and
operational cost.
Repeated scenarios
Vary accounts, order, tools, paths and timing
Baseline separation
Routine training / validation
Threshold setting
Validation subset only
Evidence integrity
Export logs where possible [14]
Missing / tamperable telemetry ≠ detector error
7.
0706 / EXPECTED RESULTS
What the comparison should reveal
Expected outcomes are hypotheses to be tested, not measured findings.
EXPECTED OUTCOMES
ANALYTICAL FOUNDATION
01
01
Linux user-activity auditing
(LUARM) [11]
02
02
All three methods on the same test set
Rule-based detection
Interpretable; potentially sensitive to legitimate variation
Behavior / correlation
Context-aware; depends on baseline, window and feature
choices
Documented evidence
Coverage, false positives, alert explanations and evidence limits
Sequence-based
masquerade detection [12]
03
03
Comparable precision,
recall and F1
ATT&CK detection guidance
and analytics [3]–[9], [13]
04
8.
0807 / TAKEAWAYS
The research in four moves
A compact view of the problem, gap, method and contribution.
01
PROBLEM
02
Privileged actions can expose sensitive local information or
stage collected data.
03
METHOD
Linux Audit → normalised events → rule / behavior /
correlation detectors → common evaluation.
GAP
No controlled comparison of three paradigms under
identical scenarios, telemetry and metrics is established in
the reviewed work.
04
CONTRIBUTION
Reproducible comparative methodology; traceable alerts;
explicit limits of host-local evidence.
SCOPE / Controlled Linux scenarios · Discovery & Collection · not all administrator roles or production environments
9.
09SOURCES
References
Core frameworks and prior work / 1 of 2
[1] MITRE, “Discovery, Tactic TA0007,” MITRE ATT&CK, accessed Oct.
2, 2026. [Online]. Available: https://attack.mitre.org/tactics/TA0007/
[7] MITRE, “Data from Local System, Technique T1005,” MITRE
ATT&CK, accessed Oct. 2, 2026. [Online]. Available:
https://attack.mitre.org/techniques/T1005/
[2] MITRE, “Collection, Tactic TA0009,” MITRE ATT&CK, accessed Oct.
2, 2026. [Online]. Available: https://attack.mitre.org/tactics/TA0009/
[8] MITRE, “Automated Collection, Technique T1119,” MITRE ATT&CK,
accessed Oct. 2, 2026. [Online]. Available:
https://attack.mitre.org/techniques/T1119/
[3] MITRE, “Process Discovery, Technique T1057,” MITRE ATT&CK,
accessed Oct. 2, 2026. [Online]. Available:
https://attack.mitre.org/techniques/T1057/
[9] MITRE, “Data Staged: Local Data Staging, Sub-technique
T1074.001,” MITRE ATT&CK, accessed Oct. 2, 2026. [Online].
Available: https://attack.mitre.org/techniques/T1074/001/
[4] MITRE, “File and Directory Discovery, Technique T1083,” MITRE
ATT&CK, accessed Oct. 2, 2026. [Online]. Available:
https://attack.mitre.org/techniques/T1083/
[10] Red Hat, “Security hardening,” RHEL 9 Documentation, accessed
Oct. 2, 2026. [Online]. Available:
https://docs.redhat.com/en/documentation/red_hat_enterprise_linux/9/
html/security_hardening/auditing-the-system_security-hardening
[5] MITRE, “System Owner/User Discovery, Technique T1033,” MITRE
ATT&CK, accessed Oct. 2, 2026. [Online]. Available:
https://attack.mitre.org/techniques/T1033/
[6] MITRE, “Permission Groups Discovery: Local Groups, Subtechnique T1069.001,” MITRE ATT&CK, accessed Oct. 2, 2026.
[Online]. Available: https://attack.mitre.org/techniques/T1069/001/
[11] G. Magklaras, S. Furnell, and M. Papadaki, “LUARM: An audit
engine for insider misuse detection,” Int. J. Digital Crime and Forensics,
vol. 3, no. 3, pp. 37–49, 2011, doi: 10.4018/jdcf.2011070103.
10.
SOURCESReferences
Prior detection and forensic research / 2 of 2
[12] J. Liu, M. Duan, W. Li, and X. Tian, “HMMs based masquerade detection for network security on with parallel computing,” Computer
Communications, vol. 156, pp. 168–173, 2020, doi: 10.1016/j.comcom.2020.03.048.
[13] MITRE, “Detection of Local Data Collection Prior to Exfiltration, Detection Strategy DET0380,” MITRE ATT&CK, accessed Oct. 2, 2026. [Online].
Available: https://attack.mitre.org/detectionstrategies/DET0380/
[14] B. Manral and G. Somani, “Establishing forensics capabilities in the presence of superuser insider threats,” Forensic Science International:
Digital Investigation, vol. 38, Art. no. 301263, 2021, doi: 10.1016/j.fsidi.2021.301263.
These are the sources cited on the slides.
10
11.
10SOURCES
AI Use Disclosure
AI Tool: ChatGPT (OpenAI)
Model: GPT-6
URL: https://chatgpt.com/
Purpose of use:
Generation of the initial presentation structure and slide content.
Improvement of academic writing and presentation wording.
Examples of prompts:
"сделай презентацию по project proposal в стиле ниу вшэ по прикрепленным критериям"
"референсы на одном слайде не помещаются что делать"
“напиши текст для выступления послайдно”
Human contribution:
The AI-generated content was reviewed, edited and adapted by the author. The author is responsible for the final content and research decisions.
Evaluation:
AI
ChatGPT helped accelerate presentation preparation and organize research materials.
12.
HSE / CYBER SECURITYThank you for
your attention!
Q&A
2026
Rustam Aliushev