ARCANUM IRL Lab — Court Filing Prompt Injection (7 Levels, 3 NPCs)

Pro Se Litigant — Historical Context & Taxonomy Classification

Origin

In October 2024, Matthew Elliott discovered that court filing systems using AI document processing were vulnerable to prompt injection through hidden text in legal filings. He demonstrated that 3-point white font — invisible to human court clerks reviewing filings — could contain instructions that AI systems would read and follow. This exposed a critical vulnerability in courts that had begun using AI to triage, summarize, or process legal documents.

The attack class is document-borne indirect prompt injection: the injection enters through a formal document (court filing) that passes through human review but contains content designed for machine consumption. The human reviewer sees nothing suspicious; the AI system reads the hidden text as a legitimate instruction.

Scenario

You are a pro se litigant (self-represented) in a court case. The court uses three AI systems:

You write court filings that these AI systems process. Your goal: use prompt injection techniques hidden in your filings to manipulate each AI system into taking unauthorized actions — from something as harmless as getting a cookie recipe to something as serious as dismissing your case and awarding you $10 million.

Taxonomy Classification

This lab maps to the Arcanum PI Taxonomy as follows:

ElementPIT CodeName
Input SurfacePIT-N-04Form (Court Filing)
Primary Intent (L1-L3)PIT-I-20Unauthorized Action Execution
Primary Intent (L4)PIT-I-19Sensitive Data Exfiltration
Primary Intent (L6-L7)PIT-I-15Jailbreak (sentence bypass)
L2 TechniquePIT-T-01Authority Framing
L3 TechniquePIT-T-14Policy Framing + Urgency (Stacked)
L4 TechniquePIT-T-15Instruction Laundering (Stacked Framing)
L5 TechniquePIT-N-05Format Confusion
L5 EvasionPIT-E-22Visual Hiding / Encoding
L6 TechniquePIT-T-19Reorientation
L7 EvasionPIT-E-07/E-08Multi-encoding (hex/ROT13/morse)

Level-by-Level Taxonomy Curriculum

LevelNPCTechniqueTaxonomy DimensionTeaching Moment
L1 — BeginnerCourtOpsDirect RequestInput: ChatUnprotected AI follows any instruction — no injection needed
L2 — BeginnerCourtOpsAuthority FramingTechnique: AuthorityClaim a role you don't have; bots trust formal documents
L3 — BeginnerCourtOpsPolicy + UrgencyTechnique: Stacked framingFirst multi-technique attack — combine two framing methods
L4 — IntermediateLegalAidInstruction LaunderingTechnique: Stacked framingWrap malicious request in legitimate context (legal research)
L5 — IntermediateLegalAidFormat ConfusionTechnique + EvasionWhite-on-white text: invisible to humans, readable by AI
L6 — AdvancedJADISReorientationTechnique: ReorientationTell the AI it misunderstood — reframe facts, don't override
L7 — EliteJADISReorientation + Multi-encodingFull taxonomy stackCombine all 4 dimensions: Intent + Technique + Evasion + Input

Concept Deep Dive: Document-Borne Indirect Prompt Injection

Court filings are formal documents that pass through human review before entering AI processing systems. The attack exploits the gap between what humans can see and what machines can read:

This is the confused deputy pattern: the AI system has authority to take actions (schedule changes, data access, case dismissal) and is confused into taking unauthorized actions by instructions it can't distinguish from legitimate ones.

Progression: From Fun to Advanced

The lab is designed as a curriculum progression:

Cross-Reference

PIT-N-06 Indirect Input
PIT-I-20 Unauthorized Action Execution
PIT-I-19 Sensitive Data Exfiltration
PIT-N-05 Form / Format Confusion
PIT-E-22 Visual Obfuscation
PIT-E-07 Encoding Evasion

Full taxonomy: arcanum-sec.github.io/arc_pi_taxonomy/