Datasets:

DimensionV
/

KR-MOB-SMISHING-v1

Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      Schema at index 1 was different: 
incident_id: string
case_type: string
theme: string
difficulty_tier: string
timeline_profile: string
start_time: timestamp[s]
end_time: timestamp[s]
platform_coverage: list<item: string>
minimum_log_sources_required: list<item: string>
full_log_sources_used: list<item: string>
noise_model_applied: struct<profile: string, background_event_ratio: double>
message_samples_used: list<item: struct<target_device: string, text: string, sender_type: string, delivery_channel: string>>
page_construction_detail_safe: struct<no_brand_copy: bool, stepper_ui_flow: list<item: string>, visual_style_notes: list<item: string>, form_risk_class: string, fields_requested: list<item: string>, detection_cues_for_human_analyst: list<item: null>>
persona: struct<user_id: string, job_title: string, risk_profile: string, devices: list<item: struct<device_id: string, platform: string, os_version: string>>>
behavioral_model_for_this_incident: struct<scenario: string, why_this_behavior: string, time_gap_rationale: string>
expected_rules: struct<must_fire: list<item: null>, should_fire: list<item: null>, optional: list<item: null>, expected_fire_path: struct<>>
event_stream: list<item: struct<_event_index: string, _phase: string, event_id: string, time: timestamp[s], log_source: string, event_type: string, severity: string, entity: struct<user_id: string, device_id: string, platform: string, os_version: string, mdm_enrolled: bool, device_fingerprint_status: string>, network: struct<carrier: string, carrier_network_type: string, roaming: bool, dst_domain: string, dst_ip: string, ttl_sec: int64, domain_age_days: int64, domain_reputation: double, resolver: string, http: struct<method: string, path: string, status_code: int64, content_type: string>>, object: struct<sender_type: string, sender_id: string, contains_url: bool, message_features: struct<urgency_language: bool, sender_verification: bool>, traffic_category: string, qtype: string, user_initiated: bool, behavior: struct<signal: string, user_interaction: string, dwell_time_ms: int64>>, correlation: struct<incident_id: string, session_id: string>, label: struct<ground_truth: string, case_type: string, attack_stage: null, activity_stage: string, theme: string>, original_source_hint: string, noise_metadata: struct<duplicate_of: string, reason: string>>>
event_stream_timeline_summary: struct<total_events: int64, malicious_events: int64, benign_context_events: int64, noise_events_included: int64, phases: list<item: struct<phase: string, time_range: string, events: int64, desc: string>>, total_elapsed_minutes: double, key_time_gaps_explained: struct<SMS_to_Web_Access: string, Web_Dwell_Time: string>>
rule_fire_verification: struct<SMISH_REDIRECT_CHAIN: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PHISH_WARNING_BYPASS: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, SENSITIVE_FORM_SEQUENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PLATFORM_DIVERGENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, POST_LURE_FINANCIAL_RISK_SIGNAL: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>>
scaling_template: struct<to_reach_200_events: struct<background_events_to_add: int64, background_composition: struct<messaging_sync: int64, email_sync: int64, push_notification: int64, software_update: int64, cdn_traffic: int64, utility_widgets: int64, mdm_checkin: int64, cloud_backup: int64>>>
vs
incident_id: string
case_type: string
theme: string
difficulty_tier: string
timeline_profile: string
start_time: timestamp[s]
end_time: timestamp[s]
platform_coverage: list<item: string>
minimum_log_sources_required: list<item: string>
full_log_sources_used: list<item: string>
noise_model_applied: struct<profile: string, background_event_ratio: double>
message_samples_used: list<item: struct<target_device: string, text: string, sender_type: string, delivery_channel: string>>
page_construction_detail_safe: struct<no_brand_copy: bool, stepper_ui_flow: list<item: string>, visual_style_notes: list<item: string>, form_risk_class: string, fields_requested: list<item: string>, detection_cues_for_human_analyst: list<item: null>>
persona: struct<user_id: string, job_title: string, risk_profile: string, devices: list<item: struct<device_id: string, platform: string, os_version: string>>>
behavioral_model_for_this_incident: struct<scenario: string, why_this_behavior: string, time_gap_rationale: string>
expected_rules: struct<must_fire: list<item: null>, should_fire: list<item: null>, optional: list<item: null>, expected_fire_path: struct<>>
event_stream: list<item: struct<_event_index: string, _phase: string, event_id: string, time: timestamp[s], log_source: string, event_type: string, severity: string, entity: struct<user_id: string, device_id: string, platform: string, os_version: string, mdm_enrolled: bool, device_fingerprint_status: string, browser: string>, network: struct<dst_domain: string, dst_ip: string, http: struct<method: string, path: string, status_code: int64, user_initiated: bool>, carrier: string, carrier_network_type: string, roaming: bool, ttl_sec: int64, domain_age_days: int64, domain_reputation: double, resolver: string, sni: string, tls_version: string>, object: struct<traffic_category: string, sender_type: string, sender_id: string, contains_url: bool, message_features: struct<urgency_language: bool, sender_verification: bool>, qtype: string, user_initiated: bool, behavior: struct<signal: string, user_interaction: string, dwell_time_ms: int64, form_fields_observed: list<item: string>>>, correlation: struct<incident_id: string, message_id: string, session_id: string>, label: struct<ground_truth: string, case_type: string, attack_stage: null, theme: string, activity_stage: string>, original_source_hint: string, noise_metadata: struct<duplicate_of: string, reason: string>>>
event_stream_timeline_summary: struct<total_events: int64, malicious_events: int64, benign_context_events: int64, noise_events_included: int64, phases: list<item: struct<phase: string, time_range: string, events: int64, desc: string>>, total_elapsed_minutes: double, key_time_gaps_explained: struct<sms_to_click: string, form_interaction: string>>
rule_fire_verification: struct<SMISH_REDIRECT_CHAIN: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PHISH_WARNING_BYPASS: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, SENSITIVE_FORM_SEQUENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PLATFORM_DIVERGENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, POST_LURE_FINANCIAL_RISK_SIGNAL: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>>
scaling_template: struct<to_reach_200_events: struct<background_events_to_add: int64, background_composition: struct<messaging_sync: int64, email_sync: int64, push_notification: int64, software_update: int64, cdn_traffic: int64, utility_widgets: int64, mdm_checkin: int64, cloud_backup: int64>>>
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 243, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 3608, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2368, in _head
                  return next(iter(self.iter(batch_size=n)))
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2573, in iter
                  for key, example in iterator:
                                      ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2060, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2082, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 572, in _iter_arrow
                  yield new_key, pa.Table.from_batches(chunks_buffer)
                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "pyarrow/table.pxi", line 5039, in pyarrow.lib.Table.from_batches
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
              pyarrow.lib.ArrowInvalid: Schema at index 1 was different: 
              incident_id: string
              case_type: string
              theme: string
              difficulty_tier: string
              timeline_profile: string
              start_time: timestamp[s]
              end_time: timestamp[s]
              platform_coverage: list<item: string>
              minimum_log_sources_required: list<item: string>
              full_log_sources_used: list<item: string>
              noise_model_applied: struct<profile: string, background_event_ratio: double>
              message_samples_used: list<item: struct<target_device: string, text: string, sender_type: string, delivery_channel: string>>
              page_construction_detail_safe: struct<no_brand_copy: bool, stepper_ui_flow: list<item: string>, visual_style_notes: list<item: string>, form_risk_class: string, fields_requested: list<item: string>, detection_cues_for_human_analyst: list<item: null>>
              persona: struct<user_id: string, job_title: string, risk_profile: string, devices: list<item: struct<device_id: string, platform: string, os_version: string>>>
              behavioral_model_for_this_incident: struct<scenario: string, why_this_behavior: string, time_gap_rationale: string>
              expected_rules: struct<must_fire: list<item: null>, should_fire: list<item: null>, optional: list<item: null>, expected_fire_path: struct<>>
              event_stream: list<item: struct<_event_index: string, _phase: string, event_id: string, time: timestamp[s], log_source: string, event_type: string, severity: string, entity: struct<user_id: string, device_id: string, platform: string, os_version: string, mdm_enrolled: bool, device_fingerprint_status: string>, network: struct<carrier: string, carrier_network_type: string, roaming: bool, dst_domain: string, dst_ip: string, ttl_sec: int64, domain_age_days: int64, domain_reputation: double, resolver: string, http: struct<method: string, path: string, status_code: int64, content_type: string>>, object: struct<sender_type: string, sender_id: string, contains_url: bool, message_features: struct<urgency_language: bool, sender_verification: bool>, traffic_category: string, qtype: string, user_initiated: bool, behavior: struct<signal: string, user_interaction: string, dwell_time_ms: int64>>, correlation: struct<incident_id: string, session_id: string>, label: struct<ground_truth: string, case_type: string, attack_stage: null, activity_stage: string, theme: string>, original_source_hint: string, noise_metadata: struct<duplicate_of: string, reason: string>>>
              event_stream_timeline_summary: struct<total_events: int64, malicious_events: int64, benign_context_events: int64, noise_events_included: int64, phases: list<item: struct<phase: string, time_range: string, events: int64, desc: string>>, total_elapsed_minutes: double, key_time_gaps_explained: struct<SMS_to_Web_Access: string, Web_Dwell_Time: string>>
              rule_fire_verification: struct<SMISH_REDIRECT_CHAIN: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PHISH_WARNING_BYPASS: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, SENSITIVE_FORM_SEQUENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PLATFORM_DIVERGENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, POST_LURE_FINANCIAL_RISK_SIGNAL: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>>
              scaling_template: struct<to_reach_200_events: struct<background_events_to_add: int64, background_composition: struct<messaging_sync: int64, email_sync: int64, push_notification: int64, software_update: int64, cdn_traffic: int64, utility_widgets: int64, mdm_checkin: int64, cloud_backup: int64>>>
              vs
              incident_id: string
              case_type: string
              theme: string
              difficulty_tier: string
              timeline_profile: string
              start_time: timestamp[s]
              end_time: timestamp[s]
              platform_coverage: list<item: string>
              minimum_log_sources_required: list<item: string>
              full_log_sources_used: list<item: string>
              noise_model_applied: struct<profile: string, background_event_ratio: double>
              message_samples_used: list<item: struct<target_device: string, text: string, sender_type: string, delivery_channel: string>>
              page_construction_detail_safe: struct<no_brand_copy: bool, stepper_ui_flow: list<item: string>, visual_style_notes: list<item: string>, form_risk_class: string, fields_requested: list<item: string>, detection_cues_for_human_analyst: list<item: null>>
              persona: struct<user_id: string, job_title: string, risk_profile: string, devices: list<item: struct<device_id: string, platform: string, os_version: string>>>
              behavioral_model_for_this_incident: struct<scenario: string, why_this_behavior: string, time_gap_rationale: string>
              expected_rules: struct<must_fire: list<item: null>, should_fire: list<item: null>, optional: list<item: null>, expected_fire_path: struct<>>
              event_stream: list<item: struct<_event_index: string, _phase: string, event_id: string, time: timestamp[s], log_source: string, event_type: string, severity: string, entity: struct<user_id: string, device_id: string, platform: string, os_version: string, mdm_enrolled: bool, device_fingerprint_status: string, browser: string>, network: struct<dst_domain: string, dst_ip: string, http: struct<method: string, path: string, status_code: int64, user_initiated: bool>, carrier: string, carrier_network_type: string, roaming: bool, ttl_sec: int64, domain_age_days: int64, domain_reputation: double, resolver: string, sni: string, tls_version: string>, object: struct<traffic_category: string, sender_type: string, sender_id: string, contains_url: bool, message_features: struct<urgency_language: bool, sender_verification: bool>, qtype: string, user_initiated: bool, behavior: struct<signal: string, user_interaction: string, dwell_time_ms: int64, form_fields_observed: list<item: string>>>, correlation: struct<incident_id: string, message_id: string, session_id: string>, label: struct<ground_truth: string, case_type: string, attack_stage: null, theme: string, activity_stage: string>, original_source_hint: string, noise_metadata: struct<duplicate_of: string, reason: string>>>
              event_stream_timeline_summary: struct<total_events: int64, malicious_events: int64, benign_context_events: int64, noise_events_included: int64, phases: list<item: struct<phase: string, time_range: string, events: int64, desc: string>>, total_elapsed_minutes: double, key_time_gaps_explained: struct<sms_to_click: string, form_interaction: string>>
              rule_fire_verification: struct<SMISH_REDIRECT_CHAIN: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PHISH_WARNING_BYPASS: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, SENSITIVE_FORM_SEQUENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, PLATFORM_DIVERGENCE: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>, POST_LURE_FINANCIAL_RISK_SIGNAL: struct<fires: bool, fire_path: string, trigger: string, within_window: bool>>
              scaling_template: struct<to_reach_200_events: struct<background_events_to_add: int64, background_composition: struct<messaging_sync: int64, email_sync: int64, push_notification: int64, software_update: int64, cdn_traffic: int64, utility_widgets: int64, mdm_checkin: int64, cloud_backup: int64>>>

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

KR-MOB-SMISHING v1

한국형 모바일 스미싱 합성 데이터셋 — SIEM 상관분석 훈련용

Korean Mobile Smishing Synthetic Dataset for SIEM Correlation Training

이 데이터셋을 만든 이유

2024년 한국의 스미싱 피해 건수는 22만 건을 넘었습니다. 전년 대비 290% 급증. 피해 금액은 1조 4천억 원에 달합니다.

택배 문자를 눌렀을 뿐인 부모님, 은행 보안 알림을 믿은 직장인, 면접 결과 문자에 속은 취업준비생. 피해자는 늘 평범한 사람들입니다.

그런데 이 공격을 막아야 할 보안팀과 AI 모델은 역설적인 상황에 놓여 있습니다. 실제 스미싱 로그는 개인정보 보호 규정과 기업 보안 정책에 묶여 공유가 불가능하고, 방어자들은 데이터 없이 직감에 의존해 탐지 규칙을 만들어야 합니다.

이 악순환을 끊기 위해 KR-MOB-SMISHING을 만들었습니다.

한국 모바일 네트워크에서 실제로 발생하는 공격 패턴을 그대로 재현하되, 개인정보 위험은 제로, 실제 피해 가능성도 제로인 합성 데이터셋입니다. SIEM 상관분석 규칙 개발, 보안 분석가 훈련, 탐지 엔지니어링 벤치마킹을 위해 설계되었습니다.

이 데이터로 훈련한 단 한 명의 보안 분석가가, 단 한 건의 실제 스미싱을 더 빨리 잡아낸다면 — 이 데이터셋을 만든 보람이 있습니다.

Why This Dataset Exists

In 2024, South Korea recorded over 220,000 smishing cases — a 290% surge from the previous year. Victims lost a combined ₩1.4 trillion, and the damage falls hardest on ordinary people: a parent who clicks a fake parcel link, an office worker who trusts a spoofed bank alert, a job seeker who opens a fraudulent interview notice.

Yet the security teams and AI models tasked with stopping these attacks face a paradox: real smishing logs are locked behind privacy regulations and corporate silos, leaving defenders to build detection rules on intuition rather than data.

We built KR-MOB-SMISHING to break that cycle.

This is a purpose-built synthetic dataset of 100 mobile smishing/phishing incidents designed for SIEM correlation rule development, security analyst training, and detection engineering benchmarking. Every incident mirrors the attack patterns actually seen on Korean mobile networks — but with zero privacy risk, zero real-world harm, and full transparency.

If one SOC analyst catches one real smishing campaign faster because they trained on this data, it was worth building.

Dataset Overview

항목	수치
Dataset Name	KR-MOB-SMISHING v1
Total Incidents	100
Avg Events / Incident	10.0
Avg Duration	81.8 min
Validation PASS Rate	100% (100/100)
Format	JSON (1 file per incident)
Language	Korean (한국어)
License	CC BY-NC 4.0

Key Features

3가지 케이스 유형: malicious(킬체인 완성), benign_lookalike(부분 공격, 사용자 중단), benign(정상 활동)
한국 실정 반영 10대 스미싱 테마
5개 SIEM 상관분석 규칙 + 사전 검증된 fire/no-fire 판정
12종 이벤트 타입 + MITRE ATT&CK for Mobile 매핑
Android/iOS 듀얼 플랫폼 커버리지
3단계 난이도 (HIGH 48%, MEDIUM 42%, LOW 10%)
RFC 2606 안전 도메인(.example), TEST-NET IP, 마스킹 전화번호
모든 SMS에 [TRAINING] 워터마크 포함

Case Type Distribution

Case Type	Count	Ratio	Description
malicious	40	40%	킬체인 완성, 피해 발생
benign_lookalike	31	31%	공격 존재하나 사용자 중단
benign	29	29%	정상 활동, 공격 없음

Difficulty Distribution

Tier	Count	Ratio	Description
DT-HIGH	48	48%	다단계 리디렉션, 경고 우회, 금융 사기
DT-MEDIUM	42	42%	중간 복잡도 공격 체인
DT-LOW	10	10%	단순 패턴, 초급 훈련용

Platform Coverage

Platform	Appearances	Ratio
Android	57	50.4%
iOS	56	49.6%

Theme Coverage (10 Scenarios)

#	Theme	Count
1	계정 보안 경고 사칭	16
2	가족/지인 사칭 긴급 송금	12
3	채용/면접 결과 사칭	12
4	인증서 갱신/만료 사칭	12
5	통신사 요금/데이터 사칭	9
6	택배/배송 실패 사칭	9
7	결제/승인 확인 사칭	9
8	모바일 금융앱 업데이트 사칭	7
9	정부지원금/환급금 사칭	7
10	공공기관 과태료/세금 사칭	6

Correlation Rule Benchmark

Rule	Fires=True	Fires=False	Fire Rate	Window
SMISH_REDIRECT_CHAIN	43	57	43%	180s
SENSITIVE_FORM_SEQUENCE	39	61	39%	600s
POST_LURE_FINANCIAL_RISK_SIGNAL	30	70	30%	900s
PHISH_WARNING_BYPASS	20	80	20%	standalone
PLATFORM_DIVERGENCE	7	93	7%	900s

Event Type Distribution

Event Type	Count	Ratio	Category
background.normal_traffic	407	39.3%	noise
dns.query	129	12.5%	network
sms.received	126	12.2%	messaging
http.request	91	8.8%	network
http.redirect	59	5.7%	network
web.risk_warning	56	5.4%	browser_security
ios.web.behavior	54	5.2%	browser_security
auth.step_up_challenge	47	4.5%	identity
android.web_submission.blocked	24	2.3%	endpoint
form.interaction	20	1.9%	web_telemetry
bank.transfer_attempt_flagged	17	1.6%	financial
device.compliance	5	0.5%	endpoint

Safety & Compliance

Check	Result
Domain TLD (.example only, RFC 2606)	PASS
IP Range (TEST-NET only)	PASS
Phone Number (masked only)	PASS
Brand Name / Logo	None (PASS)
[TRAINING] Watermark	All SMS (PASS)
Label Consistency (must_fire = malicious count)	PASS

File Structure

KR-MOB-SMISHING-v1/ ├── README.md └── incidents/ ├── inc-MOB-MAL-001.json ├── inc-MOB-BEN-002.json ├── inc-MOB-BEN-003.json └── ... (100 files)

Each JSON file is a self-contained incident with event stream, rule fire verification, scaling template, persona, and behavioral model.

Use Cases

SOC 분석가 교육 및 역량 평가
SIEM 상관분석 규칙 개발 및 튜닝
탐지 엔지니어링 CI/CD 파이프라인 테스트
퍼플팀(Purple Team) 훈련 시나리오
피싱/스미싱 탐지 학술 연구
보안 제품 벤치마킹

License

This dataset is licensed under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).

You are free to share and adapt this dataset for non-commercial purposes with appropriate attribution. For commercial use, please contact the author.

https://creativecommons.org/licenses/by-nc/4.0/

Citation

@dataset{kr_mob_smishing_v1_2026, title={KR-MOB-SMISHING v1: Korean Mobile Smishing Synthetic Dataset}, year={2026}, license={CC BY-NC 4.0}, language={Korean} }

Contact

For commercial licensing, questions, or feedback, please open an issue or contact the author. 📧 senimanyc@gmail.com

스미싱 피해는 계속 늘고 있습니다. 훈련 데이터는 턱없이 부족합니다. 이 데이터셋이 그 간극을 조금이라도 줄이는 데 기여하길 바랍니다.

Downloads last month: 7