Threat Watch

Intelligence insights on AI-enabled cyber and influence operations

01

Tradecraft

Sophistication by quarter

Sophistication (S) = autonomy + adaptivity + dependence + stage coverage + capability weight, on an absolute scale. Each dot is one operation; boxes span the middle half of a quarter and whiskers its full range. Pick a layer to compare groups: dots move to their reporting dates, boxes drop away and each group gets its own line.

Layer
Lines

* Quarter still in progress.

Components of sophistication

Click a bar to filter every chart on this page; click again to remove it.

498 operations · no filter

AAutonomy

0123456

PAdaptivity

0123

GDependence

0123

VStage coverage

0123456+

CCapability weight

0+1+2+3+4+5+6+7+

Top 12 operations by sophistication

Click a card to expand it.

AI capabilities used in operations

Operations tagged with each capability, by year reported.

243
214
163
156
107
Artifact DevelopmentSocial EngineeringIntrusion, Evasion & PersistenceReconnaissanceAttack Orchestration
2023202420252026

Models involved

Model counts depend heavily on each provider's efforts to investigate and publish incidents, so this measures disclosure as much as usage.

188 of 498 operations name a model (38%)310 name none
ModelOperations by yearCountMedian STop decile S
Claude
65919.0
GPT
624.38.4
Gemini
4358.8
DeepSeek
371418.1
Kimi
912.217.9
Grok
89.616.5
Qwen
8815.2
Llama
5610.0
GLM
415.4n/a
MiniMax
115.8n/a
2023202420252026

MITRE ATT&CK sequence

Tactics in kill-chain order. Click a tactic to list its operations.

Measure
497 of 498 operations carry at least one tactic tag (100%)1 untagged
  1. 01Reconnaissance130
  2. 02Resource Development289
  3. 03Initial Access220
  4. 04Execution84
  5. 05Persistence56
  6. 06Privilege Escalation31
  7. 07Stealth74
  8. 08Defense Impairment18
  9. 09Credential Access119
  10. 10Discovery65
  11. 11Lateral Movement34
  12. 12Collection71
  13. 13Command and Control35
  14. 14Exfiltration62
  15. 15Impact108
Fewermore recorded incidents
02

Attribution

Threat actor nations

267 of 498 operations carry an assessed origin (54%)231 unattributed

China

78operations

Median S

6.1

Top decile S

17.1

Actor type

Top actor: APT41

Profile

Russia

38operations

Median S

6.1

Top decile S

15.4

Actor type

Top actor: The Gentlemen

Profile

North Korea

40operations

Median S

4.4

Top decile S

7.8

Actor type

Top actor: BlueNoroff

Profile

Iran

22operations

Median S

3.4

Top decile S

8.9

Actor type

Top actor: APT42

Profile
03

Victims

Targeted sectors

346 of 498 operations carry a sector tag (69%)152 untagged
Median STop decile S
Individual
9957.6
Financial services
806.615.8
Government
755.618.1
Software/Hardware
58717.7
AI/ML supply chain
35411.8
Healthcare
2912.219.4
Education
259.418.4
Defense/military
234.214.8
Retail
23917.6
Energy
194.818.3
Transportation
19515.6
Telecom
175.816.4
Critical Manufacturing
169.918.2
Media
149.415.3
Hospitality
1010.418.2
Chemical
614.321.1
Water utilities
5417.9
Food/Agriculture
38.4n/a
2023202420252026

Targeted countries

306 of 498 operations name a targeted country (61%)192 name none
04

Impact

Realized harm

What the sources document actually happened to victims. One operation can carry several kinds of harm.

349 of 498 operations document realized harm (70%)149 none documented or not assessed

Financial loss by reporting date

Net money lost by victims, where the sources give a figure (56 of 110). Excludes attacker revenue and unpaid ransoms.

Net loss to victims, USD (log scale)

Sophistication Sunder 55 to 1010 to 1515+

11 more give only a range: 6 under $10k, 2 $10k to $1m, 3 $1m to $100m.

Operations are counted once each sub-incidents nested under a parent campaign are excluded. Corpus volume grows over time, so part of any upward movement reflects wider coverage rather than more activity. Sophistication is scored from each entry's description, impact and response fields.