Attack Prevalence Index

Author: ARIAtrap. Cadence: live. Data source: AgentPwn payload callbacks and TrapMyAgent fleet canaries classified into Threat Matrix techniques via the registry evidence bridge.

Methodology: /methodology/behavioral-sweep

Evidence records
5,623
30-day window
Techniques
26
distinct T-IDs
Sources
5,369
distinct honeypots
Window
Jul 9, 2026
through Aug 8, 2026

By tactic

  • Initial Access2,640
  • Privilege Escalation858
  • Persistence653
  • Reconnaissance400
  • Exfiltration338
  • Impact325
  • Credential Harvest171
  • Collection128
  • Lateral Movement110

Top techniques

  • T-2001
    Direct Prompt Injection
    Initial Access
    576
  • T-2007
    Multi-Turn Manipulation
    Initial Access
    573
  • T-2003
    Role-Play Jailbreak
    Initial Access
    453
  • T-2006
    Unicode/Encoding Bypass
    Initial Access
    383
  • T-4007
    Tool Impersonation and Squatting
    Privilege Escalation
    328
  • T-6004
    Skill/Plugin Backdoor
    Persistence
    315
  • T-2004
    Context Window Exploitation
    Initial Access
    265
  • T-6001
    Memory Injection
    Persistence
    241
  • T-8002
    HTTP Callback
    Exfiltration
    231
  • T-4001
    Capability Override
    Privilege Escalation
    224
  • T-1003
    System Prompt Extraction
    Reconnaissance
    216
  • T-9002
    Service Disruption
    Impact
    186
  • T-1002
    Tool Discovery
    Reconnaissance
    184
  • T-2005
    Tool Description Injection
    Initial Access
    181
  • T-3001
    System Prompt Credential Extraction
    Credential Harvest
    171
  • T-7004
    Memory Dump
    Collection
    128
  • T-2008
    System Prompt Boundary Bypass
    Initial Access
    110
  • T-5002
    A2A Agent Pivoting
    Lateral Movement
    110
  • T-4006
    Safety Instruction Displacement
    Privilege Escalation
    109
  • T-4002
    Admin Impersonation
    Privilege Escalation
    108

Showing top 20 of 26 techniques.

Generated Aug 8, 2026. Counts are distinct evidence rows, not raw events; the writer is idempotent so duplicate firings on a single interaction collapse to one row.

What this index measures

  • Total observed attack sessions per reporting window with a ninety-five-percent confidence interval.
  • Distribution of observations across Threat Matrix techniques. Every observation maps to exactly one primary T-NNNN code; no per-product taxonomies.
  • Partition by fleet coverage class (baseline, authenticated, dynamic, social) so the Common-Crawl-gap delta is visible.
  • Cross-property correlation rate, sessions where the same fingerprint appears on multiple fleet properties within the attribution window. Cross-property reasoning events are the highest-signal class.

See also