NEW

Start with the pressure: sales, launch, abuse, agents, data, or guardrails

Use the engagement router
aisecurityllc
Sign inScope a Launch Review
aisecurityllc
Scope a Launch Review
SecEng Adversarial RangeAI Attack Harness
by SecEng › adversarial AI security testing
9 criticalRed team run
SecEng Adversarial Range — AI Attack Harness — ACME Corp AI Assistant Platform (staging)
AI ATTACK HARNESS
Active Run
ACME Corp AI Assistant Platform
run-a7f3d9c2
GPT-4oClaude 3.5RAG (pgvector)
Categories
staging · May 21, 11:42 AM
ACME Corp AI Assistant Platform
9 critical

Scenarios Run

47

of 156 total

Total Findings

84

all severities

Critical

9

immediate action

Coverage

82%

4 frameworks

Tools Active

8

of 8 adapters

Attack Surface Radar
success rate by category
PromptInjectionRAGPoisoningDataExfilToolAbuseAgentAttacksAlignmentBypassSupplyChainModelIntegrity
Prompt Injection
87%
RAG Poisoning
62%
Data Exfil
67%
Tool Abuse
56%
Agent Attacks
54%
Alignment Bypass
34%
Supply Chain
72%
Model Integrity
34%
━━ this run╌╌ baseline
Severity Distribution
84findings
Critical
9
High
23
Medium
31
Low
21
Findings Timeline
cumulative · this run
Run start84 total
Recent Runs
9 runs
Run IDToolScen.StatusFoundCrit.Time
run-a7f3d9promptfoo18completed34411:42 AM
run-b4e1a8garak12completed27309:18 AM
run-c9d2b1PyRIT10partial15210:05 PM
run-d3f7a4AgentDojo7completed8—06:33 PM
run-e6b9c3promptfoo14completed19102:51 PM
run-f2a1c4Giskard6completed7—10:22 AM
run-g5d8b7Inspect AI5completed5—05:44 PM
run-h1e4a9NeMo Guardrails4completed4—09:31 AM
run-i7f2d5OpenAI Evals4completed3—04:05 PM
Tool Health
promptfoo
53 findings
12 runs
garak
34 findings
8 runs
PyRIT
18 findings
4 runs
AgentDojo
12 findings
3 runs
Giskard
7 findings
2 runs
Inspect AI
5 findings
2 runs
NeMo Guardrails
4 findings
1 runs
OpenAI Evals
3 findings
1 runs

Findings by tool

promptfoo
53
garak
34
PyRIT
18
AgentDojo
12
Giskard
7
Inspect AI
5
NeMo Guardrails
4
OpenAI Evals
3
run-a7f3d9c29 critical·Range: Ready·8 tools activeSecEng Adversarial Range v0.1.0

Grounded Attack Paths (APC) — from executed range findings

14 attack paths1 tools in this fixture22 evidence-graph nodes14 evidence-graph edges

Evidence from this surface, normalized and joined into the shared SecEng evidence graph. Every step traces to a specific tool’s output; grounded steps were observed or deterministically derived.

Adding a mandatory approval/control at adversarial-range:target_agent_attacks blocks 3 of 14 analyzed paths.

  • Groundedattack· conf 95%11adversarial-range
    Adversarial test inputprompt-injection
    mitre_atlas:AML.T0051owasp_llm:LLM01iso_42001:B.5.1
  • Groundedattack· conf 95%11adversarial-range
    Adversarial test inputdata-exfiltration
    mitre_atlas:AML.T0048owasp_llm:LLM02iso_42001:B.8.1
  • Groundedattack· conf 95%11adversarial-range
    Adversarial test inputdata-exfiltration
    mitre_atlas:AML.T0048owasp_llm:LLM02iso_42001:B.8.2
  • Groundedattack· conf 95%11adversarial-range
    Adversarial test inputagent-attacks
    mitre_atlas:AML.T0049owasp_llm:LLM06iso_42001:B.6.1
LegendGroundedInferredSpeculative+10 moreOpen Evidence Graph

SecEng Adversarial Range capabilities

156 pre-built attack scenarios8 tool adapters — all first-class (promptfoo, garak, PyRIT, AgentDojo, Giskard, Inspect AI, NeMo Guardrails, OpenAI Evals)ISO 42001 · EU AI Act · NIST AI RMF · MITRE ATLAS coverageSSE + WebSocket replay captureEvidence bundling with screenshot collectionReplayable regression fixturesFramework-mapped control gap registerResidual risk scorecards
Fixture: ACME Corp AI Assistant Platform · stagingRun: run-a7f3d9c2Generated: 5/21/2026SecEng Adversarial Range · v0.1.0