Prompt injection in tax technology AI

Four adversarial injection surfaces in tax technology AI

1. Tax return document injection (Intuit TurboTax AI, H&R Block AI Tax Pro)

Tax return document AI processes W-2 wage and tax statement photographs, 1099-NEC/1099-MISC/1099-K independent contractor and miscellaneous income document photographs, K-1 partnership and S-corporation income pass-through document scans, Schedule C self-employment income and expense document photographs, Form 1095-A health insurance marketplace statement scans, and supporting deduction documentation photograph uploads from Intuit TurboTax AI processing more than 40 million US tax returns per year through AI-assisted W-2 and 1099 document photograph intake, income field classification, and deduction categorisation tools that TurboTax and TurboTax Live taxpayer and tax professional users depend upon for accurate federal and state income tax liability calculations; H&R Block AI Tax Pro serving more than 20 million clients per year through AI-assisted document review, form identification, and income classification tools at H&R Block office and digital tax preparation platforms; Thomson Reuters Checkpoint AI at professional tax preparation operations processing complex Schedule C, Schedule E, Schedule F, and Form 1120/1120-S business return document images through AI-assisted professional tax research and compliance issue identification tools; and Wolters Kluwer CCH IntelliConnect AI at Fortune 500 corporate tax department operations processing complex business entity tax document images through AI-assisted corporate tax compliance and return preparation tools — extracting income classification and deduction categorisation determinations from taxpayer document photograph inputs in AI-assisted tax return preparation pipelines at filing season volumes that make individual human tax professional review of every AI-processed document photograph impracticable during peak season operations.

The adversarial injection surface is the taxpayer tax return document photograph submission pathway: TurboTax AI or H&R Block AI document photograph intake images submitted through AI-assisted income field extraction and deduction classification tools for AI tax liability calculation and return preparation. An adversarially crafted W-2 or 1099 document photograph — in which pixel perturbations applied to the Box 1 wages display region, the Box 7 nonemployee compensation indicator visual marker, or the Box 3 and Box 5 Social Security and Medicare wages display in a W-2 or 1099 document photograph cause the AI to suppress an income amount field extraction that would otherwise report taxable income to the tax return preparation AI — can create an AI-generated tax return that understates taxable income, generates an inflated refund calculation, or omits a 1099-K income disclosure that the actual document photograph evidences as reportable income meeting IRS gross income inclusion requirements under 26 USC §61. In consumer tax preparation environments where TurboTax AI processes millions of tax return document photograph uploads during peak filing season without individual human review of every AI document field extraction before the AI income classification governs the tax liability calculation, adversarial suppression of income indicators creates IRS §7206 fraudulent return and §7201 tax evasion exposure dimensions for the taxpayer whose return understates income due to AI-processed adversarially manipulated document photographs.

The IRS criminal tax liability and civil penalty consequences of adversarially suppressed income classification in tax return document AI span IRS 26 USC §7206 fraudulent return criminal liability, §7201 tax evasion criminal liability, §6663 civil fraud penalty, §6662 accuracy-related penalty, and state Department of Revenue false filing statute dimensions. IRS 26 USC §7206(1) imposes criminal liability on any person who wilfully makes or subscribes any return or document that the person does not believe to be true and correct as to every material matter — a taxpayer who submits a TurboTax AI-prepared return that understates income due to adversarially manipulated document photograph processing, and who signs the return under penalty of perjury, creates §7206(1) exposure when the taxpayer’s actual financial records disclose the understated income amounts. 26 USC §7201 imposes criminal felony liability (up to 5 years) on any person who wilfully attempts to evade or defeat any tax imposed by the Internal Revenue Code; adversarial manipulation of tax return document AI that suppresses material income amounts and generates a return that significantly understates tax liability creates §7201 tax evasion exposure when combined with affirmative acts of evasion. 26 USC §6663 imposes a 75% civil fraud penalty on the portion of any underpayment attributable to fraud; adversarially corrupted tax return document AI that generates materially understated income classifications creates §6663 civil fraud penalty exposure for the portion of tax underpayment attributable to the adversarially manipulated AI document classification. State Department of Revenue false filing statutes in California (Revenue and Taxation Code §19705), New York (Tax Law §1804), and Texas (Tax Code §171.362) impose parallel criminal and civil fraud penalty obligations for state income and franchise tax returns. Threshold: 55 for tax return document AI — reflecting the IRS §7206 fraudulent return, §7201 tax evasion, §6663 civil fraud penalty, and state DOR false filing statute dimensions of adversarially manipulated income classification.

2. Business entity and foreign filing document injection (Thomson Reuters Checkpoint AI, Wolters Kluwer CCH AI)

Business entity and foreign filing document AI processes EIN (Employer Identification Number) confirmation letter document scans, business entity formation document photographs (articles of incorporation, certificate of formation, operating agreement), foreign entity formation document images, Form 5471 Information Return of US Persons with Respect to Certain Foreign Corporations document scans, Form 5472 Information Return of a 25% Foreign-Owned US Corporation document images, FBAR FinCEN 114 foreign bank account record document photographs, FATCA Form 8938 Statement of Specified Foreign Financial Assets document scans, and FinCEN Corporate Transparency Act beneficial ownership information (BOI) reporting document images from Thomson Reuters Checkpoint AI at 750,000+ professional tax user deployments processing complex international tax and entity formation document images through AI-assisted research, compliance analysis, and filing obligation identification tools; Wolters Kluwer CCH IntelliConnect AI at 85% of Fortune 500 company tax departments processing international tax compliance document images through AI-assisted corporate tax and entity classification tools; OneSource Thomson Reuters AI at global corporate tax operations processing multinational entity structure document images; and ONESOURCE Global Trade AI at more than 150 country compliance operations processing international trade and entity formation document images — extracting entity classification, controlled foreign corporation indicator, and beneficial ownership structure determinations from business entity and foreign filing document scan inputs in AI-assisted international tax compliance and regulatory reporting pipelines.

The adversarial injection surface is the business entity formation and foreign filing document scan image submission pathway: Thomson Reuters Checkpoint AI or Wolters Kluwer CCH AI business entity document scan images submitted through AI-assisted entity classification and controlled foreign corporation indicator identification tools for AI tax compliance analysis and filing obligation generation. An adversarially crafted EIN confirmation or foreign entity formation document scan — in which pixel perturbations applied to the entity ownership percentage display region, the foreign corporation majority ownership indicator visual marker, or the beneficial ownership structure documentation display in a business entity or foreign entity formation document scan cause the AI to classify a foreign corporation meeting IRS §957 controlled foreign corporation ownership threshold criteria (50%+ US shareholder ownership) as a non-CFC entity not triggering Form 5471 filing obligations, or to classify a 25% foreign-owned US corporation as a domestic entity not subject to Form 5472 filing requirements — can suppress a CFC indicator classification that would otherwise generate a Form 5471 or Form 5472 filing obligation notification, a tax professional advisory, and an international tax compliance record. In Fortune 500 corporate tax department environments where CCH AI or Checkpoint AI processes complex multinational entity structure document images during year-end tax provision and annual return preparation without individual human review of every AI-processed entity document before the AI classification governs the international tax filing obligation determination, adversarial suppression of CFC and foreign ownership indicators creates §6038 Form 5471 penalty and FATCA §1471 withholding compliance exposure.

The IRS international tax, FBAR, FATCA, and FinCEN Corporate Transparency Act consequences of adversarially suppressed entity classification in business entity filing AI span IRS §6038 Form 5471 and Form 5472 penalty obligations, FBAR 31 USC §5314 FinCEN 114 civil and criminal penalties, FATCA 26 USC §1471 foreign financial institution withholding, FinCEN Corporate Transparency Act BOI reporting 31 USC §5336 civil penalty ($591/day), and IRS §6677 failure-to-file penalties for international information returns. IRS §6038 imposes a $10,000 penalty per year per entity for failure to file Form 5471 (US persons with interests in foreign corporations) and Form 5472 (25% foreign-owned US corporations), with additional $10,000 continuation penalties for failures exceeding 90 days after IRS notice — adversarial manipulation of Checkpoint AI or CCH AI entity classification that suppresses CFC or foreign ownership indicators creates §6038 penalty exposure when adversarially corrupted AI classifications cause tax professionals to omit required international information return filings. FBAR 31 USC §5314 requires US persons with financial interests in or signature authority over foreign bank accounts exceeding $10,000 to file FinCEN 114; FBAR civil penalties up to $10,000/year (non-willful) and $100,000 or 50% of account balance/year (willful) apply when AI-processed foreign account document photographs suppress FBAR filing obligation indicators. FATCA 26 USC §1471 imposes 30% withholding on payments to foreign financial institutions that do not comply with FATCA disclosure requirements; adversarial manipulation of AI entity classification that misidentifies US persons’ foreign financial account interests creates FATCA compliance exposure. FinCEN BOI reporting under 31 USC §5336 (Corporate Transparency Act) imposes civil penalties of $591/day and criminal penalties up to $10,000 and 2 years imprisonment for wilful BOI reporting failures — adversarially corrupted AI entity document classification that suppresses beneficial ownership structure indicators creates BOI reporting compliance exposure. Threshold: 60 for business entity filing AI — reflecting IRS §6038 international return penalties, FBAR §5314, FATCA §1471, and FinCEN CTA BOI §5336 dimensions.

3. Sales and use tax exemption certificate injection (Avalara AI, Vertex O Series AI)

Sales and use tax exemption certificate AI processes resale exemption certificate display images, manufacturing exemption certificate document photographs, agricultural exemption certificate scans, government and non-profit exemption certificate document images, direct pay permit document scans, and multi-state Streamlined Sales Tax Agreement exemption certificate display images from Avalara AI at more than 10,000 ERP integrations processing automated sales tax exemption certificate validation and nexus indicator identification through AI-assisted certificate classification and validity determination tools at e-commerce and enterprise transaction volumes; Vertex O Series AI at more than 4,000 global enterprise customers through SAP and Oracle ERP integration processing automated sales tax calculation, exemption certificate management, and economic nexus threshold monitoring through AI-assisted tax calculation and compliance tools; Sovos AI at 15,000+ customers with sales and use tax regulatory reporting obligations across 60+ countries; and OneSource Thomson Reuters AI at global corporate tax operations processing multi-jurisdictional sales and use tax compliance document images — extracting exemption certificate validity classifications and sales tax nexus indicator determinations from exemption certificate display image inputs in AI-assisted sales tax exemption management and economic nexus monitoring pipelines at transaction volumes that make individual human tax analyst review of every AI-processed exemption certificate determination impracticable for high-volume e-commerce and enterprise operations.

The adversarial injection surface is the sales and use tax exemption certificate display image submission pathway: Avalara AI or Vertex O Series AI exemption certificate display images submitted through AI-assisted certificate validity classification and nexus indicator identification tools for AI sales tax exemption determination and compliance record generation. An adversarially crafted Avalara or Vertex resale exemption certificate display — in which pixel perturbations applied to the buyer exemption status indicator display region, the certificate expiration date visual marker, or the nexus-triggering activity documentation display in a sales and use tax exemption certificate image cause the AI to classify an invalid or expired resale exemption certificate as a valid certificate meeting sales tax exemption criteria, or to classify a taxpayer’s activities as not triggering economic nexus in a state when the actual documentation evidences economic nexus-triggering transaction volumes meeting post-Wayfair South Dakota v. Wayfair 585 US ___ (2018) economic nexus threshold criteria — can suppress a nexus or exemption validity indicator that would otherwise generate a sales tax collection obligation notification, a certificate validity rejection, and a state use tax nexus compliance record. In high-volume e-commerce environments where Avalara AI processes millions of transaction-level exemption certificate validations per day without individual human tax analyst review of every AI-processed certificate determination before the AI classification governs the transaction-level sales tax collection decision, adversarial suppression of nexus or exemption validity indicators creates systematic post-Wayfair economic nexus compliance failure and state DOR audit exposure.

The post-Wayfair, Streamlined Sales Tax, and state DOR audit consequences of adversarially suppressed nexus and exemption validity classification in sales tax certificate AI span post-Wayfair South Dakota v. Wayfair 585 US ___ (2018) economic nexus compliance obligations, Streamlined Sales Tax Agreement member state audit exposure, state DOR sales and use tax audit assessment and penalty provisions, and state use tax nexus penalty and interest obligations. South Dakota v. Wayfair, 585 US ___ (2018), overruled Quill Corp. v. North Dakota’s physical presence nexus requirement and held that states may impose sales tax collection obligations on out-of-state sellers meeting economic nexus thresholds (South Dakota’s threshold: $100,000 in sales or 200 transactions annually); 45+ states have enacted post-Wayfair economic nexus standards based on the South Dakota model — adversarial manipulation of Avalara or Vertex nexus monitoring AI that suppresses economic nexus-triggering activity indicators creates systematic multi-state sales tax collection obligation failures with compound state DOR audit assessment exposure. The Streamlined Sales Tax Governing Board administers the Streamlined Sales and Use Tax Agreement (SSUTA) across 24 member states; adversarially corrupted AI exemption certificate validation that accepts invalid or manipulated SSUTA exemption certificates creates member state audit and retroactive sales tax assessment exposure across all transactions covered by adversarially validated certificates. State DOR sales and use tax audit programmes impose substantial underpayment penalties — California (10% + interest), New York (10% + 14.5% interest), and Texas (10-25% based on culpability) — and state use tax nexus penalties for out-of-state sellers failing to collect and remit required sales tax create compound multi-state penalty exposure when adversarially corrupted Avalara AI nexus classifications generate systematic collection failures across post-Wayfair economic nexus states. Threshold: 65 for sales and use tax exemption certificate AI — reflecting post-Wayfair economic nexus, SSUTA member state audit, state DOR penalty, and use tax nexus compliance dimensions.

4. Payroll tax withholding document injection (ADP AI, Ceridian Dayforce AI)

Payroll tax withholding document AI processes Form W-4 Employee’s Withholding Certificate document photographs, payroll register and pay stub display images, Form 941 Employer’s Quarterly Federal Tax Return document scans, state withholding certificate document photographs, worker classification determination documentation images (Form SS-8 and IRS determination letter scans), and payroll tax deposit confirmation document images from ADP AI payroll processing tools serving 1,000,000+ businesses and 40,000,000+ US employees processing payroll tax withholding compliance through AI-assisted payroll document classification, withholding calculation, and tax deposit management tools; Ceridian Dayforce AI at enterprise payroll operations processing payroll tax compliance document images through AI-assisted payroll tax calculation and regulatory filing tools; UKG Pro AI at enterprise payroll and workforce management operations processing payroll tax withholding compliance document images; and Wolters Kluwer CCH AI and Thomson Reuters Checkpoint AI at professional payroll tax compliance operations processing payroll tax document images through AI-assisted payroll tax research and compliance analysis tools — extracting payroll tax withholding compliance classifications and withholding shortfall indicator determinations from payroll document image inputs in AI-assisted payroll tax compliance and employer withholding obligation management pipelines at employer payroll processing volumes that make individual human payroll administrator review of every AI-processed payroll document impracticable for large employer operations.

The adversarial injection surface is the payroll tax withholding document photograph and payroll record display image submission pathway: ADP AI or Ceridian Dayforce AI payroll tax document images submitted through AI-assisted withholding compliance classification and payroll tax obligation identification tools for AI payroll tax calculation and employer withholding compliance monitoring. An adversarially crafted W-4 or payroll register document image — in which pixel perturbations applied to the federal income tax withholding amount display region, the FICA Social Security and Medicare employee withholding indicator visual marker, or the worker classification status documentation display in a payroll tax document image cause the AI to classify an employee receiving below-threshold federal income tax withholding, or a worker misclassified as an independent contractor rather than an employee under IRS common law control test criteria, as meeting payroll tax withholding compliance requirements when the actual payroll documentation evidences withholding shortfalls or worker misclassification requiring employer FICA and income tax withholding obligations — can suppress a withholding shortfall or worker misclassification indicator that would otherwise generate an employer payroll tax correction notification, a Form 941 amended return recommendation, and an IRS payroll tax deposit deficiency record. In large employer payroll environments where ADP AI or Ceridian AI processes complex multi-state payroll tax compliance determinations without individual human payroll administrator review of every AI-processed payroll document before the AI withholding classification governs the employer’s payroll tax deposit and Form 941 filing decisions, adversarial suppression of withholding shortfall indicators creates Trust Fund Recovery Penalty and worker misclassification exposure.

The IRS Trust Fund Recovery Penalty, worker misclassification, and FICA withholding consequences of adversarially suppressed withholding compliance classification in payroll tax document AI span IRS 26 USC §3102 FICA employee Social Security and Medicare tax withholding obligations, §3402 employer federal income tax withholding obligations, Trust Fund Recovery Penalty §6672 (100% personal liability penalty for responsible persons), IRS SS-8 worker status determination, and state payroll tax withholding statute penalty dimensions. IRS 26 USC §3102 requires employers to withhold employee Social Security and Medicare taxes from wages paid to employees; §3402 requires employers to withhold federal income tax from wages based on employees’ Form W-4 withholding elections — adversarial manipulation of ADP AI or Ceridian Dayforce payroll tax document classification that suppresses withholding shortfall indicators creates employer FICA and income tax withholding obligation failures that accumulate across payroll periods. The Trust Fund Recovery Penalty under 26 USC §6672 imposes 100% personal liability on any person responsible for collecting, accounting for, and paying over to the IRS withheld employee taxes (“trust fund taxes” — employee FICA and income tax withholding) who wilfully fails to collect or pay over those taxes; this penalty attaches personally to responsible corporate officers, payroll administrators, and others with control over payroll tax compliance — adversarially corrupted ADP AI or Ceridian payroll document classification that systematically suppresses withholding shortfall indicators across payroll periods creates Trust Fund Recovery Penalty personal liability exposure for responsible persons who relied on adversarially manipulated AI payroll compliance tools without additional human verification. IRS Form SS-8 worker status determination procedures assess whether workers meet IRS common law employee classification criteria based on behavioural control, financial control, and type-of-relationship factors; adversarial manipulation of worker classification document AI that suppresses employee status indicators for workers who qualify as employees under IRS common law tests creates worker misclassification back tax assessment, FICA underpayment, and employment tax audit exposure. Threshold: 60 for payroll tax document AI — reflecting IRS §3102/§3402 FICA/withholding obligations, §6672 Trust Fund Recovery Penalty personal liability, IRS SS-8 worker misclassification, and state payroll tax withholding statute dimensions.

Integration: tax technology AI image ingestion with Glyphward pre-scan

Tax technology AI image ingestion flows from TurboTax AI and H&R Block AI tax return document photograph channels, Thomson Reuters Checkpoint AI and Wolters Kluwer CCH AI business entity and foreign filing document scan interfaces, Avalara AI and Vertex O Series AI sales and use tax exemption certificate display image platforms, and ADP AI and Ceridian Dayforce AI payroll tax withholding document image processing systems into tax return income classification AI, business entity and CFC indicator extraction AI, sales tax nexus and exemption validity classification AI, and payroll withholding compliance assessment AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to tax return income classifications, entity filing obligation determinations, exemption certificate validity decisions, or payroll withholding compliance assessments:

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"

# Tax technology AI — IRS 26 USC §7206 fraudulent return (felony 3yr);
# §7201 tax evasion (felony 5yr); §6663 civil fraud penalty 75%;
# §6672 Trust Fund Recovery Penalty 100%; FBAR 31 USC §5314;
# FATCA §1471; FinCEN CTA BOI §5336; post-Wayfair economic nexus.
THRESHOLD_TAX_RETURN_DOCUMENT_AI    = 55  # TurboTax/H&R Block; §7206; §7201; §6663
THRESHOLD_BUSINESS_ENTITY_FILING_AI = 60  # CCH/Checkpoint; §6038; FBAR; FATCA; FinCEN CTA
THRESHOLD_SALES_TAX_CERTIFICATE_AI  = 65  # Avalara/Vertex; post-Wayfair; SSUTA; state DOR
THRESHOLD_PAYROLL_TAX_DOCUMENT_AI   = 60  # ADP/Ceridian; §3102; §3402; §6672 TFRP; SS-8


class TaxTechnologyAIContext(str, Enum):
    TAX_RETURN_DOCUMENT_AI    = "tax_return_document_ai"     # TurboTax, H&R Block
    BUSINESS_ENTITY_FILING_AI = "business_entity_filing_ai"  # Checkpoint, CCH, OneSource
    SALES_TAX_CERTIFICATE_AI  = "sales_tax_certificate_ai"   # Avalara, Vertex O Series
    PAYROLL_TAX_DOCUMENT_AI   = "payroll_tax_document_ai"    # ADP, Ceridian, UKG


def threshold_for(context: TaxTechnologyAIContext) -> int:
    mapping = {
        TaxTechnologyAIContext.TAX_RETURN_DOCUMENT_AI:    THRESHOLD_TAX_RETURN_DOCUMENT_AI,
        TaxTechnologyAIContext.BUSINESS_ENTITY_FILING_AI: THRESHOLD_BUSINESS_ENTITY_FILING_AI,
        TaxTechnologyAIContext.SALES_TAX_CERTIFICATE_AI:  THRESHOLD_SALES_TAX_CERTIFICATE_AI,
        TaxTechnologyAIContext.PAYROLL_TAX_DOCUMENT_AI:   THRESHOLD_PAYROLL_TAX_DOCUMENT_AI,
    }
    return mapping[context]


async def scan_tax_technology_ai_image(
    image_path: str | Path,
    context: TaxTechnologyAIContext,
    taxpayer_entity_hash: str,       # SHA-256 of taxpayer EIN, SSN hash, or entity identifier
    filing_ref: str,                 # e.g. "2025-1040-44821", "EIN-XX-XXXXXXX-5471-2025"
    tax_processing_session_id: str,  # document batch, filing season session, payroll period ID
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a tax technology AI image for adversarial injection payloads before forwarding
    to tax return income classification, business entity filing obligation extraction,
    sales tax exemption certificate validity classification, or payroll tax withholding
    compliance assessment AI systems.

    Raises AdversarialTaxTechnologyAIImageError if score meets threshold:
      - TAX_RETURN_DOCUMENT_AI:    threshold 55; §7206 fraudulent return; §7201 tax evasion
      - BUSINESS_ENTITY_FILING_AI: threshold 60; §6038 intl returns; FBAR; FATCA; FinCEN CTA
      - SALES_TAX_CERTIFICATE_AI:  threshold 65; post-Wayfair nexus; SSUTA; state DOR audit
      - PAYROLL_TAX_DOCUMENT_AI:   threshold 60; §3102/§3402 FICA; §6672 TFRP; SS-8
    """
    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": {
                "tax_technology_context":     context.value,
                "taxpayer_entity_hash":       taxpayer_entity_hash,
                "filing_ref":                 filing_ref,
                "tax_processing_session_id":  tax_processing_session_id,
                "client_scan_id":             client_scan_id,
                "image_sha256":               image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "taxpayer_entity_hash":       taxpayer_entity_hash,
        "filing_ref":                 filing_ref,
        "tax_processing_session_id":  tax_processing_session_id,
        "tax_technology_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_tax_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialTaxTechnologyAIImageError(
            f"Tax technology AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"entity={taxpayer_entity_hash} ref={filing_ref}"
        )
    return result


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


class AdversarialTaxTechnologyAIImageError(Exception):
    """Raised when a tax technology AI image exceeds the adversarial injection threshold."""
    pass

Call scan_tax_technology_ai_image() with TaxTechnologyAIContext.TAX_RETURN_DOCUMENT_AI before forwarding TurboTax AI or H&R Block AI tax return document photographs to income field extraction and deduction classification AI — with filing_ref linking the Glyphward scan to the tax return record for IRS §7206 fraudulent return, §7201 tax evasion, and §6663 civil fraud penalty compliance documentation. Call with TaxTechnologyAIContext.BUSINESS_ENTITY_FILING_AI for Thomson Reuters Checkpoint AI or Wolters Kluwer CCH AI business entity and foreign filing document scan images before entity classification and CFC indicator extraction AI, with taxpayer_entity_hash for §6038 Form 5471/5472 penalty, FBAR §5314, FATCA §1471, and FinCEN CTA BOI §5336 audit trail documentation. Call with TaxTechnologyAIContext.SALES_TAX_CERTIFICATE_AI for Avalara AI or Vertex O Series AI exemption certificate display images before certificate validity and nexus classification AI, with tax_processing_session_id as the transaction batch identifier for post-Wayfair economic nexus, SSUTA member state audit, and state DOR penalty compliance documentation. Call with TaxTechnologyAIContext.PAYROLL_TAX_DOCUMENT_AI for ADP AI or Ceridian Dayforce AI payroll tax document images before withholding compliance classification AI, with taxpayer_entity_hash for IRS §3102/§3402 FICA withholding, §6672 Trust Fund Recovery Penalty personal liability, and IRS SS-8 worker misclassification audit trail. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Financial document AI prompt injection — related attack surface covering adversarial injection in financial document processing AI with SEC, GAAP, and bank regulatory dimensions applicable to corporate tax return document AI contexts.
• Fintech and payments AI prompt injection — related regulatory framework covering payment transaction AI with FinCEN AML, BSA, and OFAC sanctions dimensions applicable to FBAR and FATCA foreign financial account AI contexts.
• SOX compliance AI security prompt injection — related adversarial attack surface covering AI injection in Sarbanes-Oxley financial controls with PCAOB, SEC, and internal controls over financial reporting dimensions applicable to corporate tax technology AI deployments.
• Free tier — 10 scans/day, no card required — start scanning tax technology AI document images at development volumes before committing to a production plan.