Prompt injection in court and judicial technology AI

Four adversarial injection surfaces in court and judicial technology AI

1. Court case management document injection (Tyler Technologies Odyssey AI)

Court case management document AI processes court filing document scan images, complaint and answer document photographs, motion and brief document scans, service-of-process return document photographs, summons and subpoena document images, appellate brief filing document scans, and court docket entry document images from Tyler Technologies Odyssey Court Manager AI at more than 70% of US state court deployments processing more than 35 million case filings per year through AI-assisted document classification, filing deadline indicator extraction, service-of-process completion marker identification, and docket management input generation tools that court clerks, judicial officers, and attorneys depend upon for FRCP Rule 4 service-of-process compliance, Rule 12(b) responsive pleading deadline management, Rule 56 summary judgment filing compliance, and state CCP statutory filing deadline management at case volume scales that make individual human court clerk examination of every AI-processed filing document impracticable for high-volume state court operations; and LexisNexis AI Judicial Analytics at court administration operations processing case document images through AI-assisted court analytics and case management tools covering 160+ countries and major US state court systems. Tyler Technologies Odyssey AI’s court case management platform processes civil and criminal case filing documents through AI-assisted document classification, deadline calculation, and docket management tools that court clerks and case administrators use for case management at filing volumes where AI-assisted processing is operationally necessary for managing statutory and rule-based filing deadlines across active case dockets.

The adversarial injection surface is the court filing document scan image submission pathway: Tyler Technologies Odyssey AI case management document scan images submitted through AI-assisted filing deadline indicator extraction and service-of-process completion classification tools for AI docket management and case timeline compliance. An adversarially crafted Tyler Technologies Odyssey case management document scan — in which pixel perturbations applied to the filing date stamp display region, the proof of service completion indicator visual marker, or the responsive pleading deadline documentation display in a court filing document scan cause the AI to suppress a filing deadline or service-of-process completion indicator that would otherwise generate a deadline management workflow entry, a service-of-process confirmation record, or a case docket compliance notification — can create a case management record that fails to reflect a filed document’s existence, a completed service of process, or an approaching responsive pleading deadline. In court clerk operations where Tyler Technologies Odyssey AI processes hundreds of daily filing documents without individual clerk examination of every AI-processed document scan before the AI classification governs the case docket management workflow, adversarial suppression of filing deadline or service-of-process indicators creates due process notice failures and FRCP procedural compliance dimension with 14th Amendment constitutional significance.

The FRCP, 14th Amendment, and Rule 11 consequences of adversarially suppressed filing classification in court case management AI span FRCP Rule 4 service-of-process accuracy requirements, Rule 12(b) responsive pleading deadline management, Rule 56 summary judgment compliance, 14th Amendment due process notice and hearing rights, Rule 11 sanctions exposure, and state CCP statutory filing deadline compliance dimensions. FRCP Rule 4 requires that defendants receive proper service of process to establish personal jurisdiction and provide constitutionally adequate notice of claims against them; Mullane v. Central Hanover Bank & Trust Co., 339 US 306 (1950), established the constitutional minimum that notice must be reasonably calculated to reach interested parties — adversarial manipulation of Tyler Technologies Odyssey AI case management document classification that suppresses proof-of-service completion indicators creates Rule 4 jurisdictional and constitutional due process notice failure dimensions when courts proceed against defendants based on service records corrupted by adversarially manipulated AI classification. FRCP Rule 12(b)(5) permits defendants to challenge the court’s jurisdiction based on insufficient service of process; adversarially corrupted Odyssey AI service classification creates Rule 12(b)(5) dismissal vulnerability when parties challenge jurisdiction based on service deficiencies that adversarial AI manipulation helped create or conceal. Rule 11 of the Federal Rules of Civil Procedure imposes sanctions obligations on attorneys who make representations to the court that are not warranted by existing law or are not supported by the factual record; an attorney who files a motion or pleading that relies on adversarially corrupted Tyler Technologies Odyssey AI docket records — without reasonable inquiry into the accuracy of AI-generated case management records — creates Rule 11 sanctions exposure when the adversarially manipulated AI output caused the attorney to make materially inaccurate representations about case procedural status. Threshold: 55 for case management document AI — reflecting FRCP Rule 4 service-of-process, Rule 12(b) responsive pleading deadline, Rule 56 summary judgment, 14th Amendment due process, and Rule 11 sanctions dimensions.

2. Recidivism risk assessment display injection (COMPAS Equivant AI, Arnold Foundation PSA AI)

Recidivism risk assessment display AI processes COMPAS risk score display images, criminal history record display visualisations, prior arrest and conviction frequency indicator displays, age-at-first-arrest indicator visualisations, prior failure-to-appear record displays, violent criminal history indicator images, and supervision response history displays from Equivant/Northpointe COMPAS AI at court deployments across Wisconsin, Florida, New Jersey, Michigan, and other state court systems using COMPAS risk scores for sentencing recommendations, parole decisions, and supervision level classifications; Arnold Foundation Public Safety Assessment AI at more than 40 US jurisdictions including New Jersey, Kentucky, New Mexico, and municipalities of major US cities processing pre-trial risk assessment factor display images through AI-assisted PSA score classification tools; and Tyler Technologies Supervision AI at probation and supervision case management operations processing supervision risk classification displays through AI-assisted supervision level and intervention classification tools — extracting risk score classification determinations and criminal history indicator assessments from risk assessment display image inputs in AI-assisted judicial sentencing and supervision decision support pipelines at court case volumes that make individual judicial officer or probation officer examination of every AI-processed risk display impracticable for courts with high-volume criminal dockets.

The adversarial injection surface is the recidivism risk score display image and criminal history indicator visualisation submission pathway: COMPAS AI or Equivant recidivism risk score display images submitted through AI-assisted risk classification and criminal history indicator identification tools for AI sentencing recommendation and supervision level determination. An adversarially crafted COMPAS recidivism risk score display — in which pixel perturbations applied to the violent recidivism risk score numerical indicator display region, the prior violent crime frequency indicator visual marker, or the criminal history severity classification display in a COMPAS risk assessment visualisation cause the AI to classify a defendant with a high recidivism risk score profile — evidencing multiple prior violent offences, prior FTA events, and substance abuse history — as a low-risk defendant not meeting high-supervision or detention recommendation criteria when the actual COMPAS display documents a risk profile meeting high-risk classification thresholds — can suppress a risk indicator that would otherwise generate a high-supervision sentencing recommendation, a detention recommendation input, and a judicial sentencing guidance record. In high-volume criminal court environments where COMPAS AI or PSA AI processes risk assessment displays for dozens of daily sentencing and bail hearings without individual judicial officer pixel-level examination of every AI-processed risk display before the AI risk classification informs the sentencing or detention recommendation, adversarial suppression of risk indicators creates 8th Amendment proportionality and Equal Protection dimensions with direct liberty consequences for crime victims and the public.

The 8th Amendment, Equal Protection, and Loomis v. Wisconsin consequences of adversarially suppressed risk classification in recidivism assessment AI span Mathews v. Eldridge 424 US 319 due process balancing, Loomis v. Wisconsin 371 Wis 2d 235 (2016) COMPAS admissibility constitutional analysis, 8th Amendment proportionality requirements for sentencing, Equal Protection 14th Amendment disparate impact dimensions identified in the ProPublica 2016 COMPAS racial disparity analysis, and state sentencing guideline procedural compliance dimensions. Loomis v. Wisconsin, 371 Wis 2d 235 (2016), examined whether a sentencing court’s consideration of COMPAS recidivism risk scores violated the defendant’s due process rights under the 14th Amendment — the Wisconsin Supreme Court held that COMPAS score consideration did not violate due process provided sentencing courts do not rely on COMPAS as the determinative factor and consider COMPAS scores alongside other evidence; adversarial injection creates a distinct vulnerability not addressed by Loomis by manipulating the COMPAS display image that the AI processes to generate its risk classification, rather than challenging the COMPAS algorithm’s validity or the sentencing court’s use of the score. The 8th Amendment prohibits excessive bail and cruel and unusual punishment and requires proportionality in criminal sentencing — Solem v. Helm, 463 US 277 (1983), established that the 8th Amendment requires proportionality review of non-capital sentences; adversarially manipulated COMPAS AI that suppresses high-risk criminal history indicators and generates artificially reduced risk scores that influence sentencing decisions creates 8th Amendment proportionality dimensions when sentences are calibrated to adversarially suppressed risk inputs. Equal Protection under the 14th Amendment, as analysed in the ProPublica 2016 study finding that COMPAS incorrectly flagged Black defendants as high-risk at nearly twice the rate of white defendants, creates additional constitutional dimensions — adversarial injection that disproportionately suppresses risk indicators for specific defendant populations creates new Equal Protection disparate impact exposure dimensions beyond the training data bias dimensions that ProPublica’s analysis addressed. Threshold: 65 for recidivism risk assessment AI — reflecting Mathews v. Eldridge due process, Loomis COMPAS admissibility, 8th Amendment proportionality, Equal Protection disparate impact, and state sentencing guideline dimensions.

3. Forensic evidence document injection (Veritone IDentify AI)

Forensic evidence document AI processes crime scene photograph exhibit images, surveillance video evidence frame captures, forensic document examination images, gunshot residue and ballistic test result display photographs, digital forensic evidence screenshot images, bodycam footage frame captures, and court exhibit forensic analysis report display images from Veritone IDentify AI at more than 1,000 law enforcement agency and court deployments processing forensic video and image analysis through AI-assisted evidence identification, subject recognition, and forensic classification tools; Thomson Reuters Westlaw Edge AI at more than 1 million legal professional deployments processing legal research document images through AI-assisted case law analysis and evidence assessment tools; LexisNexis AI Judicial Analytics at court and law firm operations processing evidentiary document images through AI-assisted judicial analytics and evidence assessment tools; and law enforcement digital forensic AI tools processing digital evidence screenshot images and forensic report displays through AI-assisted digital evidence classification and case relevance identification pipelines — extracting forensic evidence indicator classifications and criminal evidence relevance determinations from forensic exhibit photograph and evidence display image inputs in AI-assisted law enforcement evidence analysis and prosecutorial evidence assessment pipelines at evidence processing volumes that make individual human investigator or prosecutor examination of every AI-processed forensic image impracticable for large-scale criminal investigations.

The adversarial injection surface is the forensic exhibit photograph and court exhibit evidence display image submission pathway: Veritone IDentify AI or law enforcement digital forensic AI forensic evidence images submitted through AI-assisted evidence indicator classification and criminal evidence relevance identification tools for AI evidence assessment and prosecutorial evidence record generation. An adversarially crafted court exhibit photograph or Veritone IDentify forensic analysis display — in which pixel perturbations applied to the criminal evidence indicator display region, the subject identification probability visual marker, or the forensic match confidence score display in a forensic evidence exhibit image cause the AI to suppress a criminal evidence indicator or forensic match that would otherwise generate a prosecutorial evidence significance flag, an investigative lead notification, and a criminal evidence record — can create a forensic AI analysis record that fails to identify critical evidence indicators that the actual exhibit photograph documents. In criminal investigation environments where Veritone IDentify AI or digital forensic AI processes hundreds of evidence images during an investigation without individual investigator or prosecutor pixel-level examination of every AI-processed evidence image before the AI classification governs the evidence significance assessment, adversarial suppression of criminal evidence indicators creates Brady v. Maryland prosecutorial disclosure and 18 USC §1519 obstruction of justice dimensions of severe constitutional and criminal consequence.

The Brady v. Maryland, Daubert, FRE Rule 702, and 18 USC §1519 consequences of adversarially suppressed evidence classification in forensic evidence AI span Brady v. Maryland 373 US 83 prosecutorial disclosure obligations, Frye v. United States 293 F 1013 (DC Cir 1923) general acceptance standard, Daubert v. Merrell Dow 509 US 579 (1993) scientific evidence reliability and relevance admissibility standard, FRE Rule 401 relevance and Rule 702 expert testimony requirements, and 18 USC §1519 obstruction of justice criminal exposure. Brady v. Maryland, 373 US 83 (1963), held that suppression by the prosecution of evidence material to guilt or punishment violates due process — evidence is material when there is a reasonable probability that its disclosure would have produced a different result; adversarial manipulation of Veritone IDentify AI that suppresses forensic evidence indicators — indicators that would have been material to the defence or would have altered the prosecution’s evidence assessment — creates Brady disclosure obligation failures when prosecutors fail to disclose forensic analysis outputs because adversarial manipulation caused the AI to suppress the material indicator from the AI-generated forensic summary. Kyles v. Whitley, 514 US 419 (1995), extended Brady obligations to evidence possessed by law enforcement investigators even if not directly possessed by the prosecutor — adversarially manipulated Veritone IDentify AI evidence analysis records held by law enforcement investigators create Kyles-Brady disclosure dimensions for prosecutors responsible for disclosing material evidence to the defence. Daubert v. Merrell Dow, 509 US 579 (1993), established that federal courts must evaluate the reliability and relevance of expert testimony under FRE Rule 702 using factors including testability, peer review, known error rate, and general acceptance; adversarially manipulated forensic AI evidence analysis that generates inaccurate evidence classifications creates Daubert reliability challenge dimensions when forensic AI outputs are offered as expert evidence in criminal proceedings. 18 USC §1519 imposes criminal liability for knowingly falsifying, concealing, covering up, or making false entries in records or documents in connection with any federal investigation; adversarial manipulation of forensic evidence AI that creates false or misleading forensic analysis records for federal criminal investigations creates §1519 obstruction exposure. Threshold: 60 for forensic evidence document AI — reflecting Brady v. Maryland prosecutorial disclosure, Daubert/Frye scientific evidence admissibility, FRE Rule 702 expert reliability, and 18 USC §1519 obstruction dimensions.

4. Pre-trial risk assessment display injection (Arnold Foundation PSA AI, bail reform AI)

Pre-trial risk assessment display AI processes Arnold Foundation Public Safety Assessment (PSA) score display images, pre-trial risk factor indicator visualisations, flight risk classification display images, new criminal activity probability indicator visualisations, prior failure-to-appear frequency display images, and pre-trial supervision recommendation display images from Arnold Foundation PSA AI at more than 40 US jurisdictions including New Jersey (state-wide PSA implementation since 2017 bail reform), Kentucky (Administrative Office of the Courts PSA deployment), New Mexico, and more than 40 municipal and county court jurisdictions across the United States processing pre-trial risk assessment factor display images through AI-assisted PSA flight risk and new criminal activity classification tools; Equivant/Northpointe pre-trial risk assessment AI at pre-trial services programme operations; and Tyler Technologies Supervision AI at pre-trial supervision case management operations processing pre-trial risk factor display images through AI-assisted supervision level classification tools — extracting pre-trial risk classification determinations and flight risk and new criminal activity probability assessments from pre-trial risk assessment display image inputs in AI-assisted bail and detention decision support pipelines at pre-trial hearing volumes that make individual judicial officer examination of every AI-processed pre-trial risk display impracticable for high-volume criminal courts.

The adversarial injection surface is the pre-trial risk assessment factor display image and PSA score visualisation submission pathway: Arnold Foundation PSA AI or bail reform AI pre-trial risk assessment display images submitted through AI-assisted flight risk and new criminal activity classification tools for AI detention recommendation and pre-trial release condition determination. An adversarially crafted Arnold Foundation PSA or bail reform AI display — in which pixel perturbations applied to the flight risk score numerical indicator display region, the new violent criminal activity probability visual marker, or the prior failure-to-appear frequency display in a pre-trial risk assessment visualisation cause the AI to classify a defendant with a high-risk PSA profile — evidencing prior FTA events, prior new criminal activity on pretrial release, and active warrant history — as a low-risk defendant not meeting detention recommendation criteria when the actual PSA display documents a risk profile meeting detention recommendation thresholds under the jurisdiction’s PSA implementation protocols — can suppress a risk indicator that would otherwise generate a pre-trial detention recommendation, a judicial officer detention finding support input, and a pre-trial risk record. In high-volume bail hearing environments where PSA AI processes pre-trial risk displays for dozens of daily initial appearance hearings without individual judicial officer pixel-level examination of every AI-processed PSA display, adversarial suppression of flight risk and new criminal activity indicators creates Bail Reform Act §3142 and 8th Amendment excessive bail dimensions with public safety and constitutional liberty consequences.

The Bail Reform Act, 8th Amendment, and Salerno v. United States consequences of adversarially suppressed risk classification in pre-trial risk assessment AI span 18 USC §3142 Bail Reform Act detention hearing standard requirements, 8th Amendment excessive bail prohibition under Stack v. Boyle 342 US 1 (1951), Salerno v. United States 481 US 739 (1987) preventive detention constitutional standards, and 14th Amendment due process dimensions of AI-assisted pre-trial detention decision support. The Bail Reform Act, 18 USC §3142, establishes the standards governing pre-trial detention in federal criminal proceedings — requiring judicial officers to consider specified factors including the nature and circumstances of the charged offence, the weight of the evidence, the defendant’s history and characteristics, and the danger to the community in making detention determinations; adversarial manipulation of Arnold Foundation PSA AI that suppresses flight risk and new criminal activity indicators creates §3142 factor consideration failures when judicial officers rely on adversarially corrupted AI risk classifications without additional independent verification of the underlying risk factors. Stack v. Boyle, 342 US 1 (1951), established that bail set at a figure higher than necessary to assure the defendant’s appearance constitutes excessive bail under the 8th Amendment; Salerno v. United States, 481 US 739 (1987), upheld the Bail Reform Act’s preventive detention provisions against 8th Amendment and Due Process challenges, finding that the government’s interest in community safety and the procedural safeguards of the Act satisfied constitutional requirements — adversarial injection into PSA AI that suppresses community safety risk indicators creates mirror-image Salerno dimensions when adversarially manipulated AI outputs generate inappropriately low risk classifications that result in release decisions inconsistent with the Bail Reform Act’s community safety mandate. The Due Process Clause of the 14th Amendment requires that state pre-trial detention proceedings provide constitutionally adequate procedural protections; Mathews v. Eldridge balancing applied to pre-trial detention AI tools requires assessment of the risk of erroneous deprivation created by adversarially manipulable risk displays used without adequate verification procedures. Threshold: 60 for pre-trial risk assessment AI — reflecting 18 USC §3142 Bail Reform Act, 8th Amendment excessive bail Stack v. Boyle, Salerno preventive detention, and 14th Amendment Mathews due process balancing dimensions.

Integration: court and judicial technology AI image ingestion with Glyphward pre-scan

Court and judicial technology AI image ingestion flows from Tyler Technologies Odyssey Court Manager AI case management document scan channels, COMPAS Equivant AI and Arnold Foundation PSA AI risk score display interfaces, Veritone IDentify AI and digital forensic AI forensic evidence image platforms, and pre-trial risk assessment AI display processing systems into court docket management AI, recidivism risk classification AI, forensic evidence indicator AI, and pre-trial detention risk AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to case management deadline records, recidivism risk sentencing recommendations, forensic evidence significance classifications, or pre-trial detention recommendation inputs:

import asyncio
import base64
import hashlib
import os
import uuid
from enum import Enum
from pathlib import Path

import httpx

GLYPHWARD_API_KEY = os.environ["GLYPHWARD_API_KEY"]
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# Court & judicial technology AI — FRCP Rule 4 service of process; Rule 12(b)/Rule 56;
# 14th Amendment due process; Mathews v. Eldridge 424 US 319;
# Brady v. Maryland 373 US 83; Daubert v. Merrell Dow 509 US 579;
# 18 USC §1519 obstruction; 8th Amendment; Bail Reform Act 18 USC §3142;
# Stack v. Boyle 342 US 1; Salerno v. United States 481 US 739.
THRESHOLD_CASE_MANAGEMENT_DOCUMENT_AI  = 55  # Tyler Odyssey; FRCP Rule 4; 14th Amend due process
THRESHOLD_RECIDIVISM_RISK_ASSESSMENT_AI = 65  # COMPAS; Loomis; 8th Amend; Equal Protection
THRESHOLD_FORENSIC_EVIDENCE_DOCUMENT_AI = 60  # Veritone; Brady; Daubert; FRE 702; §1519
THRESHOLD_PRETRIAL_RISK_ASSESSMENT_AI   = 60  # Arnold PSA; §3142 BRA; Stack; Salerno


class CourtJudicialAIContext(str, Enum):
    CASE_MANAGEMENT_DOCUMENT_AI   = "case_management_document_ai"    # Tyler Odyssey
    RECIDIVISM_RISK_ASSESSMENT_AI = "recidivism_risk_assessment_ai"  # COMPAS, Equivant
    FORENSIC_EVIDENCE_DOCUMENT_AI = "forensic_evidence_document_ai"  # Veritone IDentify
    PRETRIAL_RISK_ASSESSMENT_AI   = "pretrial_risk_assessment_ai"    # Arnold PSA, Equivant


def threshold_for(context: CourtJudicialAIContext) -> int:
    mapping = {
        CourtJudicialAIContext.CASE_MANAGEMENT_DOCUMENT_AI:   THRESHOLD_CASE_MANAGEMENT_DOCUMENT_AI,
        CourtJudicialAIContext.RECIDIVISM_RISK_ASSESSMENT_AI: THRESHOLD_RECIDIVISM_RISK_ASSESSMENT_AI,
        CourtJudicialAIContext.FORENSIC_EVIDENCE_DOCUMENT_AI: THRESHOLD_FORENSIC_EVIDENCE_DOCUMENT_AI,
        CourtJudicialAIContext.PRETRIAL_RISK_ASSESSMENT_AI:   THRESHOLD_PRETRIAL_RISK_ASSESSMENT_AI,
    }
    return mapping[context]


async def scan_court_judicial_ai_image(
    image_path: str | Path,
    context: CourtJudicialAIContext,
    court_jurisdiction_hash: str,   # SHA-256 of court jurisdiction FIPS or PACER court ID
    case_ref: str,                  # e.g. "USDC-NDCA-2026-CV-44821", "STATE-WI-2026-CR-88841"
    judicial_session_id: str,       # document scan batch, hearing session, docket entry ID
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a court or judicial technology AI image for adversarial injection payloads
    before forwarding to court case management deadline classification, recidivism
    risk score assessment, forensic evidence indicator classification, or pre-trial
    risk assessment AI systems.

    Raises AdversarialCourtJudicialAIImageError if score meets threshold:
      - CASE_MANAGEMENT_DOCUMENT_AI:   threshold 55; FRCP Rule 4; 14th Amend; Rule 11
      - RECIDIVISM_RISK_ASSESSMENT_AI: threshold 65; Loomis; 8th Amend; Equal Protection
      - FORENSIC_EVIDENCE_DOCUMENT_AI: threshold 60; Brady; Daubert; FRE 702; 18 USC §1519
      - PRETRIAL_RISK_ASSESSMENT_AI:   threshold 60; §3142 BRA; Stack; Salerno; Mathews
    """
    image_bytes     = Path(image_path).read_bytes()
    image_b64       = base64.b64encode(image_bytes).decode()
    image_sha256    = hashlib.sha256(image_bytes).hexdigest()
    client_scan_id  = str(uuid.uuid4())
    threshold       = threshold_for(context)

    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json={
            "image": image_b64,
            "source": context.value,
            "metadata": {
                "court_judicial_context":    context.value,
                "court_jurisdiction_hash":   court_jurisdiction_hash,
                "case_ref":                  case_ref,
                "judicial_session_id":       judicial_session_id,
                "client_scan_id":            client_scan_id,
                "image_sha256":              image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "court_jurisdiction_hash":  court_jurisdiction_hash,
        "case_ref":                 case_ref,
        "judicial_session_id":      judicial_session_id,
        "court_judicial_context":   context.value,
        "scan_id":                  result["scan_id"],
        "client_scan_id":           client_scan_id,
        "image_sha256":             image_sha256,
        "score":                    result["score"],
        "flagged_region":           result.get("flagged_region"),
        "threshold":                threshold,
        "action":                   "blocked" if result["score"] >= threshold else "allowed",
    }
    await write_judicial_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialCourtJudicialAIImageError(
            f"Court judicial AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"jurisdiction={court_jurisdiction_hash} ref={case_ref}"
        )
    return result


async def write_judicial_audit_record(record: dict) -> None:
    """Persist audit record to judicial technology compliance documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialCourtJudicialAIImageError(Exception):
    """Raised when a court or judicial technology AI image exceeds the adversarial injection threshold."""
    pass

Call scan_court_judicial_ai_image() with CourtJudicialAIContext.CASE_MANAGEMENT_DOCUMENT_AI before forwarding Tyler Technologies Odyssey AI case management document scan images to filing deadline indicator and service-of-process classification AI — with case_ref linking the Glyphward scan to the case record for FRCP Rule 4 service-of-process, 14th Amendment due process, and Rule 11 sanctions compliance documentation. Call with CourtJudicialAIContext.RECIDIVISM_RISK_ASSESSMENT_AI for COMPAS Equivant AI or Arnold Foundation PSA AI risk score display images before AI risk classification and sentencing recommendation generation, with court_jurisdiction_hash for Loomis v. Wisconsin admissibility, 8th Amendment proportionality, Equal Protection disparate impact, and state sentencing guideline compliance audit trail documentation. Call with CourtJudicialAIContext.FORENSIC_EVIDENCE_DOCUMENT_AI for Veritone IDentify AI forensic evidence image or court exhibit photograph images before AI evidence indicator and criminal evidence significance classification, with judicial_session_id as the evidentiary proceeding or investigation session identifier for Brady v. Maryland prosecutorial disclosure, Daubert scientific evidence admissibility, and 18 USC §1519 obstruction compliance documentation. Call with CourtJudicialAIContext.PRETRIAL_RISK_ASSESSMENT_AI for Arnold Foundation PSA AI or bail reform AI pre-trial risk display images before flight risk and new criminal activity classification AI, with case_ref for Bail Reform Act §3142 detention hearing, 8th Amendment Stack v. Boyle excessive bail, and Salerno preventive detention compliance audit trail. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Criminal justice and forensic AI prompt injection — related attack surface covering adversarial injection in criminal justice AI systems with Fourth Amendment, Brady disclosure, Frye/Daubert admissibility, and law enforcement evidence chain-of-custody dimensions.
• Government and public sector AI prompt injection — broader regulatory framework covering 42 USC §1983 civil rights, Palantir Gotham law enforcement AI, Fourth Amendment exclusionary rule, and NVRA voter rights dimensions applicable to court and judicial technology AI contexts.
• Legal and litigation AI prompt injection — related adversarial attack surface covering AI injection in legal research and litigation support AI with Westlaw, LexisNexis, and attorney professional responsibility dimensions applicable to court technology AI deployments.
• Free tier — 10 scans/day, no card required — start scanning court and judicial technology AI images at development volumes before committing to a production plan.