Prompt injection in energy and utilities field operations AI

Four adversarial injection surfaces in energy and utilities field operations AI

1. Thermal infrared inspection injection (Teledyne FLIR AI, IntelliCheck AI)

Thermal infrared inspection AI processes thermal camera image outputs from Teledyne FLIR AI thermal camera systems at electrical utility substation, transmission line, industrial manufacturing, and government facility inspection deployments; IntelliCheck AI at electrical predictive maintenance programme operations; Fluke thermal AI at industrial and commercial facility electrical inspection programme deployments; Hikvision AI thermal imaging at electrical substation, industrial process, and building safety inspection deployments; and utility predictive maintenance AI platforms processing thermographic inspection datasets for high-voltage switchgear, transformer, bus bar, and cable termination condition assessment — extracting temperature anomaly indicator and hotspot severity classifications from thermal infrared camera image inputs in AI-assisted predictive maintenance programme trigger identification and electrical equipment condition assessment pipelines, generating hotspot severity alert records, maintenance priority work order inputs, transformer and switchgear inspection schedule adjustment recommendations, and thermographic survey compliance documentation entries that utility asset management teams, electrical maintenance engineers, and NERC CIP compliance officers depend upon for NFPA 70B predictive maintenance thermographic inspection programme fulfilment, OSHA 29 CFR §1910.303 electrical safety programme compliance, and NERC CIP-007 bulk electric system physical security inspection obligation management. Teledyne FLIR AI’s thermal camera image analysis platform processes infrared thermal camera images through AI-assisted temperature anomaly classification and hotspot severity assessment tools that electrical utility maintenance programmes use for transformers, switchgear, overhead distribution lines, and substation bus bar thermal condition monitoring at inspection survey scales where AI-assisted hotspot detection across large numbers of thermal inspection images is operationally necessary for timely identification of developing electrical failures before they progress to catastrophic equipment failure events.

The adversarial injection surface is the thermal infrared camera image submission pathway: Teledyne FLIR AI or Fluke thermal AI electrical equipment inspection camera image outputs submitted through AI-assisted temperature anomaly classification and hotspot severity assessment tools for AI predictive maintenance trigger identification and work order generation. An adversarially crafted FLIR AI thermal camera image of high-voltage electrical equipment — in which pixel perturbations applied to the hotspot temperature differential display region, the thermal gradient anomaly visual indicator, or the overheating connection terminal thermal signature in a transformer or switchgear thermal camera image cause the AI to classify electrical equipment exhibiting a developing hotspot anomaly meeting NFPA 70B thermographic inspection severity classification criteria for maintenance priority action as below-threshold normal thermal profile equipment not requiring preventive maintenance scheduling when the actual thermal image documents temperature anomaly indicators meeting NFPA 70B Category 2 or Category 3 maintenance urgency criteria — can suppress a hotspot severity alert that would otherwise generate a predictive maintenance work order, an equipment removal from service recommendation, and a thermographic survey compliance documentation record. In utility predictive maintenance environments where Teledyne FLIR AI or Hikvision AI thermal imaging processes large thermographic survey datasets from annual or quarterly electrical equipment inspection campaigns without individual electrical engineer examination of every AI-classified thermal image before the AI hotspot classification governs maintenance scheduling decisions, adversarial suppression of hotspot anomaly indicators allows developing transformer or switchgear failures to go undetected and unscheduled for maintenance with catastrophic equipment failure, power outage, electrical fire, and NFPA 70B programme compliance consequences.

The NFPA 70B, OSHA §1910.303, NERC CIP-007, and utility negligence tort consequences of adversarially suppressed hotspot anomaly classification in thermal inspection AI span NFPA 70B predictive maintenance thermographic inspection programme obligations, OSHA 29 CFR §1910.303 electrical equipment safety requirements, NERC CIP-007 bulk electric system security and physical protection standards, and utility negligence tort liability dimensions. NFPA 70B (Recommended Practice for Electrical Equipment Maintenance) provides thermographic inspection standards for electrical equipment including transformer, switchgear, and panelboard inspection intervals, hotspot severity classification categories (Category 1 through 4 based on temperature differential), and recommended maintenance response urgency for each severity category — adversarial manipulation of Teledyne FLIR AI thermal inspection processing that suppresses Category 2 or Category 3 hotspot severity classifications creates NFPA 70B predictive maintenance programme inadequacy dimensions for utility and industrial facility electrical maintenance operations. OSHA 29 CFR §1910.303 requires that electrical equipment be installed and maintained in a safe condition; adversarially suppressed thermal inspection AI that causes electrical equipment maintenance programmes to miss developing failure conditions creates OSHA §1910.303 equipment maintenance obligation failures with OSHA citation and penalty exposure when suppressed hotspot anomalies progress to equipment failures causing employee injury or workplace electrical hazard conditions. NERC CIP-007-6 (Cyber and Physical Security for Operational Technology Systems) and NERC CIP-014-3 (Physical Security of Transmission Substations) impose reliability standards for bulk electric system facilities including transmission substations — adversarially manipulated thermal inspection AI that suppresses hotspot anomaly identification in substation equipment creates NERC reliability standard compliance dimensions when asset management programmes fail to maintain substation equipment in reliable operating condition. Threshold: 55 for thermal infrared inspection AI — reflecting the NFPA 70B thermographic maintenance programme, OSHA §1910.303 electrical safety, NERC CIP bulk electric system physical security, and utility equipment failure negligence tort dimensions of adversarially suppressed hotspot anomaly classification.

2. Underground infrastructure mapping injection (Exodigo AI, ESRI ArcGIS Pro AI)

Underground infrastructure mapping AI processes subsurface utility mapping display visualisations, ground-penetrating radar (GPR) signal interpretation display graphics, utility proximity conflict assessment display images, dig ticket locate request validation display outputs, and municipal utility GIS mapping display visualisations from Exodigo AI at underground utility mapping programme deployments for telecommunications companies, electric and gas utilities, water and sewer authorities, and municipal public works departments and transportation infrastructure construction programmes; ESRI ArcGIS Pro underground utility AI at municipal GIS and utility asset management programme operations; Irth Solutions/USIC AI locate management at one-call centre integration and dig ticket processing operations; 8eleven AI underground asset mapping at oil and gas pipeline and telecommunications infrastructure mapping operations; and utility GIS AI platforms processing dig ticket and locate request data for 811 call-before-you-dig damage prevention programme compliance — extracting utility type identification and proximity conflict indicator classifications from underground mapping display image inputs in AI-assisted utility proximity assessment and excavation safety clearance pipelines, generating utility proximity conflict flags, excavation safety zone classification records, gas pipe or electrical conduit proximity warning alerts, and 811 compliance documentation entries that excavation contractors, underground utility operators, and municipal public works engineers depend upon for PHMSA 49 CFR Part 192 gas transmission and distribution pipeline safety compliance, OSHA 29 CFR §1910.269 electrical utility work practices, and state 811 dig-safe damage prevention statute compliance.

The adversarial injection surface is the Exodigo AI or ESRI ArcGIS Pro underground utility mapping display visualisation submission pathway: underground utility proximity mapping display images submitted through AI-assisted utility type identification and proximity conflict assessment tools for AI excavation safety clearance determination and utility conflict warning generation. An adversarially crafted Exodigo AI underground utility mapping display — in which pixel perturbations applied to the gas pipeline proximity indicator visual marker, the electrical conduit proximity conflict display region, or the utility depth and horizontal clearance distance visualisation in an underground mapping display image cause the AI to classify a dig site exhibiting a natural gas pipeline proximity conflict within the required PHMSA Part 192 excavation safety zone distance as a below-threshold clear excavation zone not triggering utility proximity conflict warning or hand-dig requirement when the actual mapping display documents gas pipeline proximity meeting PHMSA Part 192 excavation safety zone criteria requiring hand excavation within specified distance — can suppress a utility proximity conflict warning that would otherwise generate a hand-dig safety requirement flag, a potholing confirmation requirement, and a 811 locate request compliance documentation record. In excavation contracting environments where Exodigo AI or ESRI ArcGIS Pro underground mapping AI processes dig site proximity assessments for large excavation project scopes without individual utility locator examination of every AI-generated proximity classification before the AI conflict assessment governs the excavation contractor’s dig approach decisions, adversarial suppression of gas pipeline proximity indicators creates catastrophic gas line strike, explosion, and fatality risk with PHMSA 49 CFR Part 192 pipeline safety, 811 damage prevention statute, and excavation contractor negligence liability dimensions.

The PHMSA Part 192, OSHA §1910.269, 811 dig-safe statute, and excavation contractor negligence consequences of adversarially suppressed utility proximity classification in underground mapping AI span PHMSA 49 CFR Part 192 natural gas pipeline safety regulations, OSHA 29 CFR §1910.269 electrical utility line work practices, state 811 Call Before You Dig damage prevention statutes, and excavation contractor and utility operator tort negligence liability dimensions. PHMSA 49 CFR Part 192 establishes minimum safety standards for pipeline facilities and the transportation of gas including distribution pipelines; Part 192 Subpart L imposes requirements for excavation safety including operator notification to excavation contractors of pipeline location, marking of pipeline routes, and maintenance of minimum clearance distances during excavation — adversarially corrupted Exodigo AI underground mapping processing that suppresses gas pipeline proximity indicators creates PHMSA Part 192 excavation safety programme compliance failures when contractors excavate within pipeline safety zones without the proximity alerts that would have triggered required hand-dig procedures. OSHA 29 CFR §1910.269(u) establishes electrical utility construction and maintenance work practices including minimum approach distances from energised electrical conductors and the requirement for underground line identification before excavation — adversarially suppressed Exodigo AI mapping proximity alerts for underground electrical transmission and distribution conductors create OSHA §1910.269 electrical utility work practice compliance failures. State 811 Call Before You Dig damage prevention statutes — including Texas Utilities Code §251, California Government Code §4216, and New York Industrial Code 16-3 — require excavators to submit locate requests and receive clearance before excavating; adversarially corrupted underground mapping AI that generates incorrect clearance outputs creates 811 statute violation dimensions and excavation contractor liability for gas line damage, service outage, and consequential damages. Threshold: 50 for underground infrastructure mapping AI — reflecting the PHMSA 49 CFR Part 192 gas pipeline safety, OSHA §1910.269 electrical utility work practices, 811 dig-safe excavation damage prevention, and catastrophic infrastructure damage negligence tort dimensions of adversarially suppressed utility proximity classification, with the lower threshold reflecting the catastrophic physical harm potential of undetected underground gas pipeline strike.

3. Digital twin asset condition display injection (Honeywell Forge AI, IBM Maximo AI)

Digital twin asset condition display AI processes asset degradation condition monitoring dashboard visualisations, predictive failure probability indicator display graphics, equipment health index trend display images, maintenance intervention urgency classification visualisation outputs, and asset lifecycle remaining useful life estimate display images from Honeywell Forge AI at energy sector, oil and gas refinery, and industrial manufacturing digital twin asset performance management programme deployments; IBM Maximo AI at utility, oil and gas, transportation, and manufacturing enterprise asset management programme deployments; Emerson DeltaV AI at process industry continuous process manufacturing and energy sector control system asset management operations; ABB Ability ASPECT AI at building management and industrial facility automation system asset condition monitoring operations; Bentley Systems AssetWise AI at civil infrastructure, pipeline, and power generation asset management programme deployments; and GE Vernova GridOS AI at electric grid asset performance and reliability management programme operations — extracting asset degradation indicator and predictive failure classification inputs from digital twin asset condition monitoring dashboard display image inputs in AI-assisted predictive maintenance trigger identification and asset lifecycle management pipelines, generating degradation alert records, maintenance intervention urgency priority assignments, refinery or power plant equipment outage risk assessments, and asset replacement recommendation documentation entries that reliability engineers, operations maintenance managers, and NERC CIP compliance officers depend upon for EPA SPCC 40 CFR Part 112 spill prevention programme compliance, PHMSA gas pipeline safety management, NERC CIP-014 physical security of transmission substations, and equipment failure negligence risk management.

The adversarial injection surface is the Honeywell Forge AI or IBM Maximo AI digital twin asset condition monitoring dashboard display image submission pathway: digital twin asset condition dashboard visualisation images submitted through AI-assisted asset degradation indicator classification and predictive failure probability assessment tools for AI maintenance trigger identification and reliability engineering decision support. An adversarially crafted Honeywell Forge AI digital twin asset condition dashboard display image — in which pixel perturbations applied to the asset degradation trend indicator display region, the equipment health index trajectory visual marker, or the predictive failure probability confidence interval display in a digital twin condition monitoring dashboard cause the AI to classify a monitored asset exhibiting accelerating degradation indicators meeting the maintenance intervention urgency threshold for scheduled outage or refinery unit shutdown as a below-threshold healthy asset profile not triggering maintenance priority action when the actual dashboard documents degradation trend data meeting Honeywell Forge AI’s predictive maintenance trigger classification criteria — can suppress an asset degradation alert that would otherwise generate a maintenance work order, an equipment outage scheduling recommendation, and a reliability engineering compliance documentation record. In energy sector and industrial operations environments where Honeywell Forge AI or IBM Maximo AI processes digital twin asset condition monitoring dashboards for large refinery or power generation asset portfolios without individual reliability engineer examination of every AI-generated asset health classification, adversarial suppression of degradation indicators allows advancing equipment failures to go undetected and unscheduled with EPA SPCC spill prevention, PHMSA pipeline safety, NERC reliability standard, and catastrophic equipment failure tort liability dimensions.

The EPA SPCC, PHMSA pipeline safety, NERC CIP-014, and equipment failure negligence tort consequences of adversarially suppressed asset degradation classification in digital twin asset condition AI span EPA SPCC 40 CFR Part 112 spill prevention control and countermeasure plan obligations, PHMSA gas pipeline safety integrity management requirements, NERC CIP-014 physical security of transmission substations, and refinery or power plant equipment failure negligence tort liability dimensions. EPA SPCC 40 CFR Part 112 requires facilities that store oil in quantities above threshold amounts near waters of the US to develop, implement, and periodically review Spill Prevention, Control, and Countermeasure plans that include equipment inspection and integrity testing programmes; adversarially manipulated Honeywell Forge AI or IBM Maximo AI digital twin condition monitoring that suppresses tank, pipeline, or equipment degradation indicators creates EPA SPCC plan inspection and integrity testing programme inadequacy dimensions with EPA Region enforcement authority and facility operator civil penalty exposure. PHMSA 49 CFR Part 192 Subpart O requires gas distribution operators to implement integrity management programmes for distribution pipelines in high consequence areas including corrosion control, condition monitoring, and anomaly investigation; adversarially corrupted IBM Maximo AI asset condition monitoring that suppresses pipeline degradation indicators creates PHMSA pipeline integrity management programme compliance failures with PHMSA enforcement authority and pipeline operator penalty exposure. NERC CIP-014-3 requires transmission owners and operators to perform physical security risk assessments and implement physical security plans for transmission substations whose compromise could adversely affect the reliable operation of the bulk electric system — adversarially manipulated GE Vernova GridOS AI or Honeywell Forge AI digital twin condition monitoring at bulk electric system substations creates NERC reliability standard compliance dimensions. Threshold: 60 for digital twin asset condition AI — reflecting the EPA SPCC spill prevention programme compliance, PHMSA gas pipeline integrity management, NERC CIP-014 bulk electric system physical security, and catastrophic equipment failure negligence tort dimensions of adversarially suppressed asset degradation classification.

4. Building energy audit display injection (Schneider Electric EcoStruxure AI)

Building energy audit display AI processes building energy efficiency gap analysis dashboard visualisation displays, ASHRAE 90.1 energy efficiency standard compliance analysis display graphics, ENERGY STAR certification gap assessment display images, building automation system performance deviation indicator visualisation outputs, and energy efficiency improvement opportunity priority ranking display images from Schneider Electric EcoStruxure Asset Advisor AI at commercial building, industrial facility, data centre, and healthcare campus energy management programme deployments; Johnson Controls OpenBlue AI at commercial real estate, healthcare, and educational facility building management programme operations; Siemens Desigo CC AI at commercial and industrial building automation and energy management operations; BuildingIQ AI at commercial office building and campus energy optimisation programme deployments; Enlighted IoT AI at commercial smart building energy management operations; Carrier Digital AI at commercial building HVAC and refrigeration system automation and energy management programme deployments; and ENERGY STAR Portfolio Manager AI at EPA programme energy benchmarking operations — extracting energy efficiency gap indicator and ASHRAE compliance deficit classifications from building energy audit dashboard display image inputs in AI-assisted energy efficiency optimisation and regulatory compliance analysis pipelines, generating energy efficiency gap priority alerts, ASHRAE 90.1 compliance gap remediation recommendations, ENERGY STAR certification gap assessment records, IRS 179D deduction qualification status determinations, and utility energy efficiency rebate programme compliance documentation entries that energy management engineers, sustainability officers, and commercial real estate asset managers depend upon for ASHRAE 90.1 building energy efficiency standard compliance, ENERGY STAR certification programme requirement fulfilment, IRS 179D commercial buildings energy efficiency deduction eligibility maintenance, and state utility energy efficiency programme compliance.

The adversarial injection surface is the Schneider Electric EcoStruxure AI or Johnson Controls OpenBlue AI building energy audit dashboard display image submission pathway: building energy audit efficiency analysis dashboard visualisation images submitted through AI-assisted energy efficiency gap indicator classification and ASHRAE/ENERGY STAR compliance deficit assessment tools for AI energy optimisation priority determination and regulatory compliance status classification. An adversarially crafted Schneider EcoStruxure building energy audit dashboard display image — in which pixel perturbations applied to the energy efficiency gap magnitude indicator display region, the ASHRAE 90.1 baseline compliance deficit visual marker, or the ENERGY STAR certification gap percentage display in a building energy audit dashboard visualisation cause the AI to classify a building energy performance dataset exhibiting significant ASHRAE 90.1 or ENERGY STAR compliance gaps meeting the energy efficiency improvement programme trigger threshold as a below-threshold compliant energy performance profile not requiring efficiency improvement programme action when the actual dashboard documents energy efficiency gaps meeting ASHRAE 90.1 compliance deficit criteria or ENERGY STAR certification eligibility failure threshold — can suppress an energy efficiency gap alert that would otherwise generate an energy efficiency improvement programme priority action, an ASHRAE 90.1 compliance remediation recommendation, and an IRS 179D deduction qualification status assessment. In commercial real estate asset management environments where Schneider Electric EcoStruxure AI or Johnson Controls OpenBlue AI processes building energy audit dashboards for large commercial property portfolios without individual energy manager examination of every AI-generated efficiency classification before the AI gap identification governs energy programme investment decisions and IRS 179D deduction eligibility determinations, adversarial suppression of efficiency gap indicators allows ASHRAE compliance deficits and ENERGY STAR certification gaps to go unaddressed with IRS 179D deduction disqualification and state utility programme compliance consequences.

The IRS 179D, ASHRAE 90.1, ENERGY STAR certification, and state utility energy efficiency programme consequences of adversarially suppressed efficiency gap classification in building energy audit AI span IRS 26 USC §179D energy efficiency commercial buildings deduction eligibility, ASHRAE 90.1 building energy efficiency standard compliance, ENERGY STAR certification programme qualification requirements, and state utility energy efficiency programme rebate and incentive compliance dimensions. IRS 26 USC §179D allows commercial building owners and designers of government-owned buildings to deduct the cost of energy efficiency improvements to lighting, HVAC, and building envelope systems that achieve specified energy efficiency improvements over ASHRAE 90.1 baseline standards — the deduction ranges from $0.50 to $5.00 per square foot (indexed for inflation under the Inflation Reduction Act of 2022 amendments) based on the percentage energy reduction achieved relative to the ASHRAE 90.1 baseline. Adversarially corrupted Schneider EcoStruxure AI or BuildingIQ AI building energy audit processing that suppresses energy efficiency gap identification causes building energy management programmes to miss efficiency improvement opportunities that would have qualified for §179D deduction, creating IRS deduction opportunity loss dimensions — and where AI-generated clean energy audit reports are used to support §179D deduction claims that later fail IRS examination because actual building performance does not meet efficiency improvement thresholds, adversarially corrupted energy AI creates §179D deduction recapture and accuracy-related penalty exposure. ASHRAE 90.1 (Energy Standard for Sites and Buildings Except Low-Rise Residential Buildings) is the model energy code adopted by most US state building codes and IECC energy code frameworks — adversarially suppressed EcoStruxure AI efficiency gap classification that allows buildings to maintain ASHRAE 90.1 compliance gaps creates state building code and energy code enforcement dimensions. ENERGY STAR Portfolio Manager AI benchmarking certification at the EPA requires buildings to demonstrate actual measured energy performance in the top 25% of similar building types nationally — adversarially corrupted energy audit AI that suppresses performance gap identification creates ENERGY STAR certification revocation risk when buildings are certified based on adversarially optimistic audit AI assessments that mask actual performance deficiencies. Threshold: 60 for building energy audit AI — reflecting the IRS 26 USC §179D energy efficiency deduction eligibility, ASHRAE 90.1 building code compliance, ENERGY STAR certification programme qualification, and state utility programme compliance dimensions of adversarially suppressed efficiency gap classification.

Integration: energy and utilities field operations AI image ingestion with Glyphward pre-scan

Energy and utilities field operations AI image ingestion flows from Teledyne FLIR AI and Fluke thermal infrared camera image channels, Exodigo AI and ESRI ArcGIS Pro underground utility mapping display interfaces, Honeywell Forge AI and IBM Maximo AI digital twin asset condition monitoring dashboard display platforms, and Schneider Electric EcoStruxure AI and Johnson Controls OpenBlue AI building energy audit dashboard display systems into predictive maintenance hotspot severity classification AI, excavation safety utility proximity conflict assessment AI, asset degradation indicator and maintenance trigger classification AI, and building energy efficiency gap identification and ASHRAE/ENERGY STAR compliance analysis AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to thermographic maintenance work orders, underground excavation safety clearances, asset degradation alerts, or energy efficiency compliance certifications:

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"

# Energy & utilities field operations AI — NFPA 70B thermographic inspection;
# OSHA 29 CFR §1910.303 electrical safety; NERC CIP-007/-014 bulk electric system;
# OSHA 29 CFR §1910.269 electrical utility work practices;
# PHMSA 49 CFR Part 192 natural gas pipeline safety; 811 dig-safe statutes;
# EPA SPCC 40 CFR Part 112; IRS 26 USC §179D energy efficiency deduction.
THRESHOLD_THERMAL_INSPECTION_AI  = 55  # FLIR/Fluke; NFPA 70B; OSHA §1910.303; NERC CIP
THRESHOLD_UNDERGROUND_MAPPING_AI = 50  # Exodigo/ESRI; PHMSA Part 192; OSHA §1910.269; 811
THRESHOLD_DIGITAL_TWIN_ASSET_AI  = 60  # Honeywell/Maximo; EPA SPCC; PHMSA; NERC CIP-014
THRESHOLD_BUILDING_ENERGY_AUDIT_AI = 60 # EcoStruxure/OpenBlue; IRS §179D; ASHRAE 90.1


class EnergyUtilitiesAIContext(str, Enum):
    THERMAL_INSPECTION_AI   = "thermal_inspection_ai"    # Teledyne FLIR, Fluke, Hikvision
    UNDERGROUND_MAPPING_AI  = "underground_mapping_ai"   # Exodigo, ESRI, USIC
    DIGITAL_TWIN_ASSET_AI   = "digital_twin_asset_ai"    # Honeywell Forge, IBM Maximo
    BUILDING_ENERGY_AUDIT_AI = "building_energy_audit_ai" # EcoStruxure, OpenBlue, BuildingIQ


def threshold_for(context: EnergyUtilitiesAIContext) -> int:
    mapping = {
        EnergyUtilitiesAIContext.THERMAL_INSPECTION_AI:    THRESHOLD_THERMAL_INSPECTION_AI,
        EnergyUtilitiesAIContext.UNDERGROUND_MAPPING_AI:   THRESHOLD_UNDERGROUND_MAPPING_AI,
        EnergyUtilitiesAIContext.DIGITAL_TWIN_ASSET_AI:    THRESHOLD_DIGITAL_TWIN_ASSET_AI,
        EnergyUtilitiesAIContext.BUILDING_ENERGY_AUDIT_AI: THRESHOLD_BUILDING_ENERGY_AUDIT_AI,
    }
    return mapping[context]


async def scan_energy_utilities_ai_image(
    image_path: str | Path,
    context: EnergyUtilitiesAIContext,
    operator_id_hash: str,          # SHA-256 of utility operator or facility identifier
    asset_or_site_ref: str,         # e.g. "FLIR-XFMR-2026-44821", "EXODIGO-SITE-88841"
    inspection_session_id: str,     # thermographic survey run, dig ticket batch, or audit period
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan an energy or utilities field operations AI image for adversarial injection payloads
    before forwarding to thermal infrared hotspot severity classification, underground utility
    proximity conflict assessment, digital twin asset degradation trigger identification, or
    building energy efficiency gap analysis AI systems.

    Raises AdversarialEnergyUtilitiesAIImageError if score meets threshold:
      - THERMAL_INSPECTION_AI:    threshold 55; NFPA 70B; OSHA §1910.303; NERC CIP
      - UNDERGROUND_MAPPING_AI:   threshold 50; PHMSA Part 192; OSHA §1910.269; 811
      - DIGITAL_TWIN_ASSET_AI:    threshold 60; EPA SPCC; PHMSA; NERC CIP-014
      - BUILDING_ENERGY_AUDIT_AI: threshold 60; IRS 26 USC §179D; ASHRAE 90.1; ENERGY STAR
    """
    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": {
                "energy_utilities_context": context.value,
                "operator_id_hash":         operator_id_hash,
                "asset_or_site_ref":        asset_or_site_ref,
                "inspection_session_id":    inspection_session_id,
                "client_scan_id":           client_scan_id,
                "image_sha256":             image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "operator_id_hash":        operator_id_hash,
        "asset_or_site_ref":       asset_or_site_ref,
        "inspection_session_id":   inspection_session_id,
        "energy_utilities_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_energy_utilities_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialEnergyUtilitiesAIImageError(
            f"Energy/utilities AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"operator={operator_id_hash} ref={asset_or_site_ref}"
        )
    return result


async def write_energy_utilities_audit_record(record: dict) -> None:
    """Persist audit record to utility compliance and field operations documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialEnergyUtilitiesAIImageError(Exception):
    """Raised when an energy or utilities field operations AI image exceeds the adversarial injection threshold."""
    pass

Call scan_energy_utilities_ai_image() with EnergyUtilitiesAIContext.THERMAL_INSPECTION_AI before forwarding Teledyne FLIR AI or Fluke thermal AI electrical equipment inspection camera images to temperature anomaly and hotspot severity classification AI — with asset_or_site_ref linking the Glyphward scan to the equipment inspection record for NFPA 70B thermographic maintenance programme compliance, OSHA 29 CFR §1910.303 electrical safety, and NERC CIP bulk electric system physical security audit documentation. Call with EnergyUtilitiesAIContext.UNDERGROUND_MAPPING_AI for Exodigo AI or ESRI ArcGIS Pro underground utility proximity mapping display images before AI utility type identification and proximity conflict assessment, with inspection_session_id as the dig ticket or locate request identifier for PHMSA 49 CFR Part 192 gas pipeline safety, OSHA §1910.269 electrical utility work practices, and 811 dig-safe damage prevention statute compliance documentation — the 50 threshold reflects the catastrophic physical harm potential of undetected underground gas pipeline proximity. Call with EnergyUtilitiesAIContext.DIGITAL_TWIN_ASSET_AI for Honeywell Forge AI, IBM Maximo AI, or GE Vernova GridOS AI digital twin asset condition monitoring dashboard display images before AI asset degradation indicator classification and predictive failure probability assessment, with operator_id_hash for EPA SPCC 40 CFR Part 112 spill prevention programme, PHMSA pipeline integrity management, and NERC CIP-014 transmission substation physical security compliance audit trail. Call with EnergyUtilitiesAIContext.BUILDING_ENERGY_AUDIT_AI for Schneider EcoStruxure AI or Johnson Controls OpenBlue AI building energy audit dashboard display images before AI energy efficiency gap identification and ASHRAE/ENERGY STAR compliance deficit classification, with inspection_session_id as the energy audit period identifier for IRS 26 USC §179D deduction eligibility, ASHRAE 90.1 building code compliance, and ENERGY STAR certification programme compliance audit documentation. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Indirect prompt injection via image — foundational attack pattern underlying all four energy and utilities field operations AI injection surfaces; covers how adversarial pixel-level perturbations cause AI misclassification without detectable visual artefacts at human review resolution.
• Industrial IoT AI prompt injection — related attack surface analysis covering IIoT sensor data AI, SCADA system AI, process control monitoring AI, and industrial automation AI adversarial injection with OSHA Process Safety Management, IEC 62443, and EPA RMP regulatory dimensions applicable to energy and utilities field operations AI contexts.
• Vision-language model security — technical architecture of adversarial image attacks against vision-language models including pixel perturbation classes applicable to thermal infrared camera image AI, underground mapping visualisation AI, digital twin dashboard AI, and building energy audit display AI processing pipelines.
• Free tier — 10 scans/day, no card required — start scanning energy and utilities field operations AI images at development volumes before committing to a production plan.