Prompt injection in ESG and sustainability reporting AI

Four adversarial injection surfaces in ESG and sustainability reporting AI

1. Scope 1/2/3 emissions document injection (Persefoni AI, Watershed AI)

Scope 1/2/3 emissions document AI processes utility bill photograph uploads, natural gas and fuel invoice document scan images, refrigerant recharge record photographs, employee business travel receipt and mileage log scan images, supply chain tier-1 supplier emissions disclosure document photographs, GHG measurement and monitoring report display images, and Scope 3 purchased goods and services emissions factor documentation photographs from Persefoni AI at major bank, asset manager, and enterprise carbon accounting deployments serving 1,000+ companies processing utility bill and fuel invoice photographs through AI-assisted Scope 1/2/3 GHG quantity extraction and carbon inventory classification tools; Watershed AI at Fortune 500 enterprise carbon management operations including Airbnb, Stripe, and Microsoft processing energy use and emissions evidence document images through AI-assisted GHG accounting and reporting pipelines; Measurabl AI at commercial real estate ESG operations tracking more than 20 billion square feet across 150+ countries processing utility bill and energy consumption document images through AI-assisted energy use intensity and GHG emissions quantity extraction tools; and Greenly AI at 2,000+ company SMB carbon accounting operations processing business emissions document photographs through AI-assisted Scope 1/2/3 emissions inventory tools — extracting GHG emissions quantity classifications and carbon accounting indicator determinations from emissions document photograph inputs in AI-assisted corporate carbon inventory and sustainability disclosure pipelines at enterprise emissions documentation volumes that make individual human sustainability analyst review of every AI-processed utility bill or fuel invoice photograph impracticable for large enterprise carbon accounting operations.

The adversarial injection surface is the utility bill, fuel invoice, or GHG measurement report photograph image submission pathway: Persefoni AI or Watershed AI emissions document photograph images submitted through AI-assisted GHG emissions quantity extraction and Scope classification tools for AI carbon inventory record generation and sustainability disclosure input. An adversarially crafted utility bill, fuel invoice, or GHG measurement report photograph — in which pixel perturbations applied to the energy consumption quantity display region, the fuel volume or CO2 equivalent indicator visual marker, or the measurement period emissions total display in an emissions document photograph cause the AI to suppress an emissions quantity indicator or Scope categorisation that would otherwise generate a carbon inventory entry, a GHG Protocol-compliant Scope 1/2/3 emissions record, and a sustainability disclosure input — can create a carbon accounting record that understates corporate GHG emissions quantities relevant to SEC climate disclosure materiality determinations and EU CSRD mandatory GHG reporting obligations. In enterprise sustainability operations where Persefoni AI or Watershed AI processes hundreds of utility bill and fuel invoice photographs per reporting period without individual human sustainability analyst examination of every AI-processed emissions document before the AI emissions quantity classification governs the corporate carbon inventory, adversarial suppression of emissions quantity indicators creates SEC Reg S-K Item 1500 material climate disclosure accuracy and EU CSRD ESRS E1 mandatory GHG reporting accuracy dimensions.

The SEC climate disclosure, EU CSRD, California SB 253, and GHG Protocol consequences of adversarially suppressed emissions quantity classification in emissions document AI span SEC climate disclosure rules 17 CFR §229 Reg S-K Item 1500 material climate risk disclosure obligations, EU CSRD ESRS E1 mandatory Scope 1/2/3 GHG disclosure requirements (mandatory 2025 for large EU companies), California SB 253 Climate Corporate Data Accountability Act (mandatory Scope 1/2/3 reporting for California-operating companies with $1B+ revenue), GHG Protocol Corporate Standard measurement accuracy requirements, and TCFD recommendations framework climate-related financial disclosure dimensions. SEC Reg S-K Item 1500 requires public companies to disclose material climate-related risks, Scope 1 and Scope 2 GHG emissions (and Scope 3 where material or included in emissions targets), and climate-related financial metrics in their annual reports; adversarial manipulation of Persefoni AI or Watershed AI emissions document classification that suppresses material emissions quantity indicators creates SEC Reg S-K Item 1500 material disclosure accuracy failures and SEC Rule 10b-5 material misstatement dimensions when publicly traded companies rely on adversarially corrupted AI carbon accounting outputs for their securities disclosure filings. EU CSRD ESRS E1 requires mandatory third-party assured disclosure of Scope 1/2/3 GHG emissions under the GHG Protocol Corporate Standard beginning in 2025 for large EU companies and 2026 for listed SMEs; adversarially corrupted Persefoni AI or Watershed AI emissions classification creates CSRD ESRS E1 GHG reporting accuracy failure dimensions with EU regulatory enforcement exposure. California SB 253 requires California-operating companies with $1 billion or more in annual revenues to publicly disclose Scope 1, 2, and 3 GHG emissions beginning in 2026; adversarial manipulation of emissions document AI creates SB 253 disclosure accuracy dimensions with California Air Resources Board enforcement exposure. Threshold: 60 for emissions document AI — reflecting SEC Reg S-K Item 1500 material climate disclosure, EU CSRD ESRS E1 mandatory GHG reporting, California SB 253, GHG Protocol accuracy, and TCFD recommendations dimensions.

2. ESG rating display injection (MSCI ESG AI, Bloomberg ESG AI, S&P Global ESG AI)

ESG rating display AI processes MSCI ESG rating report display visualisation images, Bloomberg ESG score and data layer display screenshots, S&P Global Trucost physical risk and carbon data display images, Refinitiv ESG score display screenshots, Sustainalytics ESG risk rating display images, and ISS ESG rating and proxy voting recommendation display images from MSCI ESG AI rating more than 12,000 securities ESG performance for 6,000+ institutional investor clients including Barclays, BlackRock, and CalPERS through AI-assisted ESG risk indicator extraction from company disclosure document images and MSCI ESG rating display processing tools; Bloomberg ESG AI delivering ESG data through the Bloomberg Terminal to more than 350,000 subscribers through AI-assisted ESG indicator extraction from company disclosure data and ESG rating display images; S&P Global ESG AI providing Trucost physical risk and carbon data through AI-assisted climate risk and ESG performance indicator extraction tools integrated into S&P ESG Index and S&P Global Ratings; Refinitiv ESG AI at 500+ institutional client deployments processing company ESG disclosure and rating display images through AI-assisted ESG indicator extraction tools; and Clarity AI at 200+ institutional investor client deployments processing ESG regulatory analytics and rating display images through AI-assisted sustainability and ESG compliance classification tools — extracting ESG risk indicator classifications and investment-grade ESG performance assessments from ESG rating display visualisation inputs in AI-assisted institutional investment research and portfolio ESG assessment pipelines at investment analysis volumes that make individual analyst review of every AI-processed ESG rating display impracticable for large institutional investment operations.

The adversarial injection surface is the MSCI ESG rating report display, Bloomberg ESG score display, or S&P Global ESG rating display image submission pathway: MSCI ESG AI, Bloomberg ESG AI, or S&P Global ESG AI rating display images submitted through AI-assisted ESG risk indicator extraction and investment-grade ESG assessment tools for AI ESG investment analysis record generation and portfolio ESG classification. An adversarially crafted MSCI ESG rating report display or Bloomberg ESG score display — in which pixel perturbations applied to the ESG rating category indicator display region, the governance or social risk score visual marker, or the environmental risk rating downgrade signal display in an ESG rating visualisation cause the AI to suppress a material ESG risk indicator or rating downgrade signal that would otherwise generate an investment risk flag, a portfolio ESG exclusion trigger, and an institutional investor ESG due diligence record — can create an AI ESG investment analysis record that fails to identify material ESG risks that the actual rating display documents. In institutional investment operations where MSCI ESG AI or Bloomberg ESG AI processes thousands of security-level ESG rating displays per day without individual analyst pixel-level examination of every AI-processed ESG rating display before the AI classification governs the portfolio ESG assessment, adversarial suppression of material ESG risk indicators creates SEC Rule 10b-5 material misstatement, ERISA fiduciary duty, and EU SFDR Article 8/9 sustainable product disclosure dimensions.

The SEC Rule 10b-5, ERISA fiduciary, EU SFDR, and ISSB IFRS S1/S2 consequences of adversarially suppressed ESG risk indicator classification in ESG rating display AI span SEC Rule 10b-5 17 CFR §240.10b-5 material misstatement liability, DOL PTE 2020-02 ERISA fiduciary ESG investment obligation, EU SFDR Article 8/9 sustainable financial product disclosure requirements, ISSB IFRS S1/S2 sustainability-related financial disclosure standards, and EU Taxonomy Article 8 mandatory sustainability reporting dimensions. SEC Rule 10b-5 prohibits material misstatements or omissions in connection with the purchase or sale of any security; investment managers who rely on adversarially corrupted MSCI ESG AI or Bloomberg ESG AI rating displays to make securities investment decisions and who disclose ESG investment criteria in their fund prospectuses, shareholder communications, or Form ADV investment adviser disclosures without disclosing material ESG risks that adversarial manipulation of their AI tools caused them to miss create SEC Rule 10b-5 material misstatement dimensions when those omissions are material to fund investors. DOL PTE 2020-02 and ERISA fiduciary standards require investment advisers and plan fiduciaries managing ERISA-covered pension and 401(k) assets to act solely in the economic interest of plan participants; adversarial manipulation of MSCI ESG AI or Bloomberg ESG AI rating display classification that suppresses material ESG risk indicators affecting the economic performance of investment holdings creates ERISA fiduciary investment prudence obligation dimensions when plan fiduciaries incorporate ESG risk analysis generated by adversarially corrupted AI tools in their investment due diligence. EU SFDR Articles 8 and 9 impose disclosure obligations on financial market participants offering sustainable investment products — Article 8 products must disclose how environmental and social characteristics are promoted; Article 9 products must disclose sustainable investment objectives and principal adverse impact indicators; adversarially corrupted MSCI ESG AI or Bloomberg ESG AI rating displays that suppress material ESG risk indicators create EU SFDR Article 8/9 product disclosure accuracy failures with European Securities and Markets Authority (ESMA) enforcement exposure. Threshold: 55 for ESG rating display AI — reflecting SEC Rule 10b-5 material misstatement, DOL PTE 2020-02 ERISA fiduciary, EU SFDR Article 8/9 product disclosure, ISSB IFRS S1/S2 sustainability disclosure, and EU Taxonomy Article 8 dimensions.

3. Supply chain ESG audit document injection (supply chain due diligence AI)

Supply chain ESG audit document AI processes supplier ESG audit report document scan images, third-party social compliance audit report photographs, factory inspection report display images, supplier environmental management system certification document scans, supplier labour practice documentation photographs, forced labour and child labour risk assessment report display images, and conflict minerals due diligence report document scan images from enterprise supply chain ESG AI platforms at major multinational corporation supply chain due diligence operations processing supplier audit document images through AI-assisted supplier ESG compliance indicator extraction and supply chain risk classification tools; EcoAct AI (Schneider Electric) at large enterprise ESG consulting operations processing supply chain emissions and sustainability audit document images through AI-assisted supply chain ESG assessment tools; S&P Global ESG AI at supply chain risk assessment operations processing supplier ESG performance and audit document images through AI-assisted supply chain ESG classification tools; Refinitiv ESG AI and Clarity AI at institutional investor supply chain ESG due diligence operations processing supplier audit report images through AI-assisted supply chain human rights and environmental risk classification tools; and ONESOURCE Thomson Reuters AI and Sovos AI at global corporate compliance operations processing international supply chain due diligence and trade compliance document images through AI-assisted multi-jurisdictional supply chain compliance assessment tools — extracting supply chain ESG compliance indicator classifications and human rights and environmental violation risk determinations from supplier audit document scan image inputs in AI-assisted supply chain due diligence and mandatory reporting pipelines.

The adversarial injection surface is the supplier ESG audit report document scan and third-party audit document photograph image submission pathway: supply chain ESG AI or S&P Global ESG AI supplier audit document scan images submitted through AI-assisted supplier compliance indicator classification and human rights and environmental violation risk identification tools for AI supply chain due diligence record generation and mandatory reporting input. An adversarially crafted supplier ESG audit report or factory inspection document scan — in which pixel perturbations applied to the forced labour risk indicator display region, the environmental violation finding visual marker, or the safety and working conditions non-compliance documentation display in a supplier audit document scan cause the AI to classify a supplier audit report documenting significant human rights violations or environmental compliance failures as a compliant supplier audit not meeting violation flag criteria when the actual document scan evidences supply chain violations meeting CSDDD or LkSG mandatory due diligence reporting thresholds — can suppress a supply chain violation indicator that would otherwise generate a supplier remediation requirement, a mandatory due diligence finding record, and an annual supply chain transparency statement disclosure input. In enterprise supply chain due diligence operations where supply chain ESG AI processes hundreds of supplier audit document scans per reporting cycle without individual human sustainability analyst examination of every AI-processed audit document before the AI compliance indicator classification governs the supply chain due diligence assessment, adversarial suppression of human rights and environmental violation indicators creates CSDDD and LkSG mandatory due diligence failure dimensions.

The EU CSDDD, German LkSG, UK Modern Slavery Act, Dodd-Frank §1502, and California SCA consequences of adversarially suppressed supply chain violation classification in supply chain ESG audit AI span EU Corporate Sustainability Due Diligence Directive (CSDDD) mandatory human rights and environmental due diligence obligations, German Supply Chain Due Diligence Act (LkSG/Lieferkettensorgfaltspflichtengesetz) mandatory due diligence requirements for companies with 1,000+ Germany-based employees, UK Modern Slavery Act §54 supply chain transparency statement obligations, Dodd-Frank §1502 conflict minerals 17 CFR §240.13p-1 SEC filing requirements, and California Transparency in Supply Chains Act (SCA) §1714.43 mandatory disclosure obligations. EU CSDDD requires large EU companies and non-EU companies with significant EU operations to conduct mandatory human rights and environmental due diligence across their supply chains, including risk identification, risk prevention and mitigation measures, and complaint mechanism implementation — adversarial manipulation of supply chain ESG AI that suppresses human rights or environmental violation indicators in supplier audit classifications creates CSDDD mandatory due diligence failure dimensions with EU member state supervisory authority enforcement exposure and civil liability for supply chain harm victims. The German LkSG requires companies with 1,000+ employees in Germany (expanded from 3,000+ in 2024) to conduct annual human rights and environmental due diligence including risk analysis, risk prevention measures, and annual due diligence reporting to BAFA (Federal Office for Economic Affairs and Export Control); adversarially corrupted AI supply chain audit classification creates LkSG annual due diligence report accuracy failures and BAFA administrative fine exposure up to 2% of global annual revenue for LkSG violations. UK Modern Slavery Act §54 requires commercial organisations with annual turnover of £36 million or more that supply goods or services in the UK to publish an annual transparency statement disclosing steps taken to ensure slavery and human trafficking are not present in their supply chains; adversarial manipulation of supply chain ESG AI that suppresses forced labour indicators in supplier audit classifications creates MSA §54 transparency statement accuracy obligation dimensions. Threshold: 60 for supply chain ESG audit AI — reflecting EU CSDDD human rights due diligence, German LkSG mandatory due diligence, UK Modern Slavery Act §54 transparency, Dodd-Frank §1502 conflict minerals, and California SCA §1714.43 dimensions.

4. Green building and LEED certification document injection (LEED AI, ENERGY STAR AI)

Green building and LEED certification document AI processes LEED v4.1 Building Design + Construction (BD+C) certification packet document photographs, ENERGY STAR Score and Portfolio Manager performance report display images, ASHRAE 90.1 energy code compliance documentation scan photographs, energy audit report display images, utility benchmarking report document photographs, commissioning report document scan images, and IRS §179D Energy Efficient Commercial Buildings Deduction tax certification document photographs from Measurabl AI at commercial real estate ESG operations tracking more than 20 billion square feet across 150+ countries processing energy audit and building performance document images through AI-assisted energy use intensity and LEED/ENERGY STAR performance classification tools; CBRE ESG AI, JLL ESG AI, and Cushman & Wakefield ESG AI at commercial real estate ESG performance operations processing green building certification document images through AI-assisted energy performance and certification compliance classification tools; energy management AI platforms at large commercial building portfolio operations processing utility benchmarking report and energy audit document images through AI-assisted energy performance gap identification and LEED/ENERGY STAR compliance monitoring tools; and building certification consulting AI platforms at LEED certification and ENERGY STAR certification support operations processing LEED scorecard and certification packet document images through AI-assisted LEED point compliance and certification eligibility classification tools — extracting energy performance indicator classifications and LEED/ENERGY STAR certification compliance determinations from green building certification document photograph inputs in AI-assisted building sustainability compliance and tax incentive eligibility assessment pipelines.

The adversarial injection surface is the LEED certification packet, energy audit report, or ENERGY STAR scorecard photograph image submission pathway: Measurabl AI or CBRE ESG AI green building certification document photograph images submitted through AI-assisted energy performance gap indicator extraction and LEED/ENERGY STAR compliance classification tools for AI green building certification compliance determination and IRS §179D deduction eligibility assessment. An adversarially crafted LEED certification packet, energy audit report, or ENERGY STAR Portfolio Manager performance report display — in which pixel perturbations applied to the energy use intensity performance gap indicator display region, the ENERGY STAR score threshold visual marker, or the LEED scorecard point compliance documentation display in a green building certification document photograph cause the AI to suppress an energy performance gap indicator or LEED point compliance shortfall that would otherwise generate a certification ineligibility notification, a deduction eligibility disqualification assessment, and a building energy performance compliance record — can create a green building certification AI record that overstates energy performance compliance and LEED certification eligibility. In commercial real estate portfolio operations where Measurabl AI or building ESG AI processes hundreds of green building certification document photographs per reporting period without individual human energy analyst examination of every AI-processed certification document before the AI performance classification governs the building certification eligibility and IRS §179D deduction eligibility determination, adversarial suppression of energy performance gap indicators creates IRS deduction eligibility fraud and LEED certification programme compliance failure dimensions.

The IRS §179D, Fannie Mae Green MBS, ASHRAE 90.1, state utility incentive, and LEED v4.1 consequences of adversarially suppressed energy performance classification in green building certification AI span IRS 26 USC §179D Energy Efficient Commercial Buildings Deduction eligibility requirements (up to $5.65/sqft for qualifying energy-efficient commercial buildings), Fannie Mae Green Mortgage Backed Securities energy performance standards, ASHRAE 90.1 energy code compliance certification requirements, state utility incentive programme performance verification requirements, and LEED v4.1 BD+C scorecard point compliance and certification tier eligibility requirements. IRS 26 USC §179D allows building owners and, for government-owned buildings, the designer of energy-efficient property to claim deductions of up to $5.65/sqft (2026 inflation-adjusted) for energy-efficient commercial building property installed as part of the interior lighting systems, HVAC systems, or building envelope; §179D eligibility requires certification by a qualified engineer or contractor that the property meets the applicable ASHRAE 90.1 energy cost reduction standards — adversarial manipulation of building energy performance AI that suppresses energy performance gap indicators and generates inflated energy efficiency compliance assessments creates IRS §179D deduction eligibility fraud dimensions when the adversarially corrupted AI classifications support deduction eligibility certifications for buildings that do not meet §179D energy reduction thresholds. Fannie Mae Green MBS programme standards require that green-labelled multifamily mortgage pools demonstrate verifiable energy and water efficiency performance improvements; adversarially corrupted Measurabl AI or building ESG AI energy performance classifications that suppress performance gap indicators create Fannie Mae Green MBS programme eligibility accuracy dimensions when adversarially inflated AI energy performance assessments support green mortgage pool eligibility determinations. ASHRAE 90.1 energy code compliance is a prerequisite for §179D deduction eligibility and LEED certification under multiple LEED v4.1 credit pathways; adversarially corrupted AI compliance assessment creates ASHRAE 90.1 certification accuracy failures with compound §179D and LEED eligibility consequences. State utility incentive programmes — including California’s PG&E, SCE, and SDG&E energy efficiency incentive programmes, New York’s NYSERDA programmes, and the DOE Weatherization Assistance Programme — require verified energy performance documentation for incentive eligibility; adversarially suppressed energy performance gap indicators in AI energy audit classification create incentive programme eligibility fraud exposure. Threshold: 65 for green building certification AI — reflecting IRS §179D deduction eligibility fraud, Fannie Mae Green MBS performance standards, ASHRAE 90.1 energy code compliance, state utility incentive programme verification, and LEED v4.1 BD+C scorecard point compliance dimensions.

Integration: ESG and sustainability reporting AI image ingestion with Glyphward pre-scan

ESG and sustainability reporting AI image ingestion flows from Persefoni AI and Watershed AI Scope 1/2/3 emissions document photograph channels, MSCI ESG AI and Bloomberg ESG AI ESG rating display visualisation interfaces, supply chain ESG AI and S&P Global ESG AI supplier audit document scan platforms, and Measurabl AI and CBRE ESG AI green building certification document photograph processing systems into emissions quantity classification AI, ESG risk indicator extraction AI, supply chain violation identification AI, and green building energy performance classification AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to carbon inventory records, ESG investment rating classifications, supply chain due diligence findings, or green building certification eligibility determinations:

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"

# ESG & sustainability reporting AI — SEC Reg S-K Item 1500 climate disclosure;
# EU CSRD ESRS E1 mandatory GHG reporting; California SB 253;
# SEC Rule 10b-5 material misstatement; DOL PTE 2020-02 ERISA fiduciary;
# EU SFDR Article 8/9; EU CSDDD; German LkSG; UK Modern Slavery Act §54;
# IRS §179D Energy Efficient Commercial Buildings Deduction; LEED v4.1.
THRESHOLD_EMISSIONS_DOCUMENT_AI   = 60  # Persefoni/Watershed; SEC Reg S-K; CSRD ESRS E1
THRESHOLD_ESG_RATING_DISPLAY_AI   = 55  # MSCI/Bloomberg/S&P; Rule 10b-5; ERISA; SFDR
THRESHOLD_SUPPLY_CHAIN_AUDIT_AI   = 60  # supply chain ESG; CSDDD; LkSG; UK MSA §54
THRESHOLD_GREEN_BUILDING_CERT_AI  = 65  # Measurabl/CBRE; §179D; Fannie Mae; LEED v4.1


class ESGSustainabilityAIContext(str, Enum):
    EMISSIONS_DOCUMENT_AI  = "emissions_document_ai"   # Persefoni, Watershed, Measurabl
    ESG_RATING_DISPLAY_AI  = "esg_rating_display_ai"   # MSCI ESG, Bloomberg ESG, S&P
    SUPPLY_CHAIN_AUDIT_AI  = "supply_chain_audit_ai"   # S&P Global, EcoAct, Refinitiv
    GREEN_BUILDING_CERT_AI = "green_building_cert_ai"  # Measurabl, CBRE, JLL ESG


def threshold_for(context: ESGSustainabilityAIContext) -> int:
    mapping = {
        ESGSustainabilityAIContext.EMISSIONS_DOCUMENT_AI:  THRESHOLD_EMISSIONS_DOCUMENT_AI,
        ESGSustainabilityAIContext.ESG_RATING_DISPLAY_AI:  THRESHOLD_ESG_RATING_DISPLAY_AI,
        ESGSustainabilityAIContext.SUPPLY_CHAIN_AUDIT_AI:  THRESHOLD_SUPPLY_CHAIN_AUDIT_AI,
        ESGSustainabilityAIContext.GREEN_BUILDING_CERT_AI: THRESHOLD_GREEN_BUILDING_CERT_AI,
    }
    return mapping[context]


async def scan_esg_sustainability_ai_image(
    image_path: str | Path,
    context: ESGSustainabilityAIContext,
    reporting_entity_hash: str,   # SHA-256 of company EIN, LEI, or organisation identifier
    disclosure_ref: str,          # e.g. "ESG-10K-2026-44821", "LEED-CERT-2026-88841"
    esg_session_id: str,          # reporting period batch, rating analysis session, audit ID
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan an ESG or sustainability reporting AI image for adversarial injection payloads
    before forwarding to Scope 1/2/3 emissions quantity classification, ESG risk rating
    indicator extraction, supply chain human rights and environmental violation identification,
    or green building energy performance certification compliance AI systems.

    Raises AdversarialESGSustainabilityAIImageError if score meets threshold:
      - EMISSIONS_DOCUMENT_AI:  threshold 60; SEC Reg S-K Item 1500; CSRD ESRS E1; SB 253
      - ESG_RATING_DISPLAY_AI:  threshold 55; SEC Rule 10b-5; ERISA; EU SFDR Art 8/9
      - SUPPLY_CHAIN_AUDIT_AI:  threshold 60; CSDDD; LkSG; UK Modern Slavery Act §54
      - GREEN_BUILDING_CERT_AI: threshold 65; IRS §179D; Fannie Mae Green MBS; LEED v4.1
    """
    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": {
                "esg_sustainability_context": context.value,
                "reporting_entity_hash":      reporting_entity_hash,
                "disclosure_ref":             disclosure_ref,
                "esg_session_id":             esg_session_id,
                "client_scan_id":             client_scan_id,
                "image_sha256":               image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "reporting_entity_hash":      reporting_entity_hash,
        "disclosure_ref":             disclosure_ref,
        "esg_session_id":             esg_session_id,
        "esg_sustainability_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_esg_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialESGSustainabilityAIImageError(
            f"ESG sustainability AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"entity={reporting_entity_hash} ref={disclosure_ref}"
        )
    return result


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


class AdversarialESGSustainabilityAIImageError(Exception):
    """Raised when an ESG or sustainability reporting AI image exceeds the adversarial injection threshold."""
    pass

Call scan_esg_sustainability_ai_image() with ESGSustainabilityAIContext.EMISSIONS_DOCUMENT_AI before forwarding Persefoni AI or Watershed AI Scope 1/2/3 emissions utility bill and fuel invoice photographs to GHG quantity extraction and carbon inventory classification AI — with disclosure_ref linking the Glyphward scan to the carbon reporting record for SEC Reg S-K Item 1500 material climate disclosure, EU CSRD ESRS E1 mandatory GHG reporting, and California SB 253 compliance documentation. Call with ESGSustainabilityAIContext.ESG_RATING_DISPLAY_AI for MSCI ESG AI, Bloomberg ESG AI, or S&P Global ESG AI rating display visualisation images before ESG risk indicator extraction and investment analysis classification AI, with reporting_entity_hash for SEC Rule 10b-5 material misstatement, DOL PTE 2020-02 ERISA fiduciary, and EU SFDR Article 8/9 sustainable product disclosure compliance audit trail documentation. Call with ESGSustainabilityAIContext.SUPPLY_CHAIN_AUDIT_AI for supplier ESG audit report document scan images before supply chain human rights and environmental violation indicator classification AI, with esg_session_id as the audit reporting cycle identifier for EU CSDDD mandatory due diligence, German LkSG BAFA reporting, UK Modern Slavery Act §54 transparency statement, and Dodd-Frank §1502 conflict minerals compliance documentation. Call with ESGSustainabilityAIContext.GREEN_BUILDING_CERT_AI for LEED certification packet, energy audit report, or ENERGY STAR scorecard photograph images before energy performance gap indicator and LEED compliance classification AI, with disclosure_ref for IRS §179D deduction eligibility, Fannie Mae Green MBS performance standards, ASHRAE 90.1 energy code compliance, and LEED v4.1 BD+C scorecard point compliance audit trail. Get early access

Coverage matrix

Frequently asked questions

Further reading

• SOX compliance AI security prompt injection — related attack surface covering adversarial injection in Sarbanes-Oxley financial controls AI with PCAOB, SEC, and internal controls over financial reporting dimensions applicable to ESG and sustainability disclosure AI contexts where climate disclosure and financial reporting intersect.
• Financial document AI prompt injection — related regulatory framework covering financial document processing AI with SEC, GAAP, and bank regulatory dimensions applicable to ESG materiality assessment AI that processes financial and non-financial disclosure document images.
• Energy and utilities field operations AI prompt injection — related adversarial attack surface covering AI injection in energy sector operational AI with FERC, NERC, and utility regulatory dimensions applicable to emissions measurement and Scope 1 direct emissions document AI contexts.
• Free tier — 10 scans/day, no card required — start scanning ESG and sustainability reporting AI document images at development volumes before committing to a production plan.