Prompt injection in smart grid and power distribution AI

Four adversarial injection surfaces in smart grid and power distribution AI

1. Substation equipment condition photograph injection (GE Vernova GridIQ AI, Siemens Energy Management AI, ABB Ability AI)

Substation equipment condition AI processes thermal infrared inspection photographs of power transformers, switchgear assemblies, disconnect switches, and circuit breakers, as well as visible-spectrum condition survey images of transformer oil containment areas, bushing condition photographs, surge arrester inspection images, and arc flash event indicator photographs, submitted through AI-assisted predictive maintenance, asset health management, and substation condition monitoring tools that extract transformer thermal hotspot temperature differentials, switchgear contact degradation classifications, oil leak severity grades, bushing insulation condition ratings, and arc flash precursor indicator flags from substation equipment image inputs, generating maintenance priority assignments, emergency de-energisation recommendations, and NERC CIP-014-3 physical security-relevant condition assessments that govern when transmission substation assets require immediate out-of-service action versus deferred scheduled maintenance. GE Vernova GridIQ AI processes substation thermal inspection photographs and equipment condition survey images through AI-assisted asset health management and predictive maintenance tools at more than 100 utility operators globally, including transmission system operators and integrated utilities across North America, Europe, and Asia Pacific whose substations serve as critical nodes in transmission networks carrying thousands of megawatts of bulk electric power. Siemens Energy Management AI processes substation condition survey photographs and SCADA-integrated substation equipment status images through AI-assisted maintenance planning and grid state monitoring tools at transmission system operators worldwide, with AI-generated condition assessments informing transmission constraint management decisions and outage scheduling determinations that affect grid reliability at the interconnection level. ABB Ability AI processes power transformer thermal monitoring images, protection relay display photographs, and switchgear condition inspection images through AI-assisted energy management and predictive maintenance tools at SCADA/EMS operators managing 700+ MW of generation and transmission capacity, with AI condition grade outputs directly informing N-1 contingency planning and capital maintenance programme prioritisation decisions at integrated utility operators across North America and Europe.

The adversarial injection surface is the transformer thermal inspection photograph, switchgear condition survey image, and arc flash precursor indicator photograph submission pathway: substation equipment condition images submitted through GE Vernova GridIQ AI, Siemens Energy Management AI, or ABB Ability AI substation monitoring interfaces for AI thermal degradation detection, oil leak classification, and arc flash precursor identification. An adversarially crafted substation equipment condition photograph — in which pixel perturbations applied to the thermal hotspot indicator region, transformer oil discolouration display, or arc tracking carbonisation pattern on a substation inspection image cause the AI to classify the equipment as within-normal operating parameters when the actual image documents a thermal degradation condition meeting NERC CIP-014-3 transmission substation physical security-relevant asset condition threshold or a transformer thermal exceedance requiring immediate de-energisation — can suppress a condition alert that would otherwise mandate an emergency maintenance intervention or equipment outage, allowing a degraded transformer or switchgear assembly to remain in service while thermal runaway, oil fire, or arc flash risk continues to develop. In high-voltage transmission substations where power transformers represent decades-long capital investments and single-unit failures can remove hundreds of megawatts of transmission capacity from service, adversarial suppression of a thermal degradation classification in AI-assisted condition monitoring defers the required maintenance action while the thermal damage mechanism progresses, potentially converting a manageable planned maintenance outage into an unplanned catastrophic transformer failure with grid reliability and physical safety consequences that extend far beyond the substation fence. The adversarial perturbation required to suppress a thermal hotspot classification in a substation condition monitoring AI system is structurally indistinguishable from ordinary thermal image noise or camera calibration artefacts at human visual inspection resolution — an adversarially manipulated substation thermal photograph presented to a substation engineer for visual review will not reveal the pixel-level perturbation pattern that causes AI misclassification, because the perturbation is optimised for AI model decision boundaries, not for human visual pattern detection thresholds.

The regulatory consequences of adversarially suppressed substation equipment condition detection in grid AI span NERC CIP, FERC Order 693, and physical safety dimensions of exceptional severity. NERC CIP-014-3 (Physical Security) requires transmission owners and transmission operators to assess the physical security risk of transmission stations and substations that, if rendered inoperable or damaged, could result in widespread instability, uncontrolled separation, or cascading within an interconnection; adversarial AI manipulation that suppresses a transformer condition alert at a CIP-014-3 high-impact transmission substation creates a physical security management failure with NERC civil penalty exposure beginning at $1M per day per violation under FERC Order 693 NERC CIP enforcement. NERC CIP-007-6 (Systems Security Management) requires responsible entities to implement security management controls for cyber assets within electronic security perimeters including BES Cyber Systems used for grid monitoring and control; adversarial manipulation of AI-assisted substation condition monitoring tools that are integrated with SCADA and EMS Cyber Systems creates CIP-007-6 security management compliance concerns where the AI monitoring tool's image ingestion boundary constitutes an unprotected attack surface within the electronic security perimeter. FERC Order 693 (Mandatory Reliability Standards for the Bulk-Power System) mandates compliance with NERC reliability standards including CIP standards by all applicable entities; FERC enforcement of NERC CIP violations arising from adversarially manipulated substation monitoring AI operates under FERC's civil penalty authority with penalty amounts that can reach $1M+ per day per violation and can extend over the full period during which the manipulated AI monitoring condition persisted. Beyond regulatory enforcement, a transformer oil fire or arc flash event at a high-voltage transmission substation caused by adversarially suppressed AI condition monitoring creates physical safety consequences — electrical arc flash injuries, fire suppression personnel exposure, and transformer oil environmental release — that generate OSHA recordable event reporting obligations and potential criminal liability under state environmental law. Threshold: 50 for substation equipment condition AI — the strictest threshold, reflecting life-safety and bulk electric system reliability primacy.

2. Smart meter display data injection (Itron AI, Landis+Gyr AI)

Smart meter display AI processes photographs of advanced metering infrastructure (AMI) display panels, meter reading confirmation screens, tamper indicator display images, and meter enclosure condition photographs submitted through AI-assisted revenue protection, energy theft detection, and tariff compliance monitoring tools that extract abnormal consumption pattern flags, tamper seal breach classifications, meter reversal indicators, bypass wiring condition grades, and foreign connection anomaly markers from meter display and condition image inputs, generating energy theft investigation referral priorities, tariff compliance enforcement flags, and FERC Order 2222 distributed energy resource (DER) settlement anomaly alerts that govern when utility revenue protection field teams are dispatched for site investigation, when meter data is excluded from DER aggregation settlement, and when utility billing exception workflows are triggered for potential revenue loss remediation. Itron AI is deployed at more than 8,000 utility customers globally — including investor-owned utilities, municipal utilities, and electric cooperatives across North America, Europe, and Asia Pacific — processing smart meter display photographs and AMI head-end display screenshots through AI-assisted revenue protection analytics, demand response verification, and distributed energy resource monitoring tools that serve as the primary automated detection layer for metering anomalies and energy theft patterns across utility customer populations numbering in the millions. Landis+Gyr AI processes more than 600 million meter readings per day across its global deployed AMI base, with AI-assisted anomaly detection and tamper classification tools processing meter display photographs submitted through utility field operations applications, remote meter management portals, and DER aggregation settlement verification platforms at utility operators whose combined metering installations span hundreds of millions of customer endpoints across multiple grid interconnections. Oracle Utilities AI augments meter display AI with billing data visualisation processing and load forecasting validation tools that generate billing exception alerts and demand forecast confidence scores from visualisation display screenshots, with AI-generated billing exception classifications governing revenue protection workflow initiation at large investor-owned utility operators whose billing systems process tens of millions of customer accounts.

The adversarial injection surface is the smart meter display photograph, tamper indicator image, and meter enclosure condition photograph submission pathway: meter display images submitted through Itron AI revenue protection interfaces or Landis+Gyr AI AMI analytics platforms for AI abnormal consumption classification, tamper indicator detection, and energy theft anomaly identification. An adversarially crafted smart meter display photograph — in which pixel perturbations applied to the tamper seal indicator region, consumption register display, or bypass wiring evidence area on a meter enclosure inspection image cause the AI to classify the meter installation as compliant when the actual image documents a tamper seal breach, register manipulation evidence, or external bypass wiring condition consistent with systematic energy theft — can suppress a revenue protection flag that would otherwise trigger a utility field investigation referral, allowing energy theft to continue undetected for billing cycles that may extend months before the next meter inspection event. In utility service territories where AI-assisted revenue protection tools operate as the primary automated detection mechanism for metering anomalies across millions of AMI customer endpoints, adversarial suppression of a tamper indicator classification or abnormal consumption pattern flag in a single AI processing cycle defers the required field investigation while the underlying metering anomaly continues to accumulate revenue loss exposure and, in cases of deliberate bypass wiring, creates physical electrical safety hazards — exposed conductor connections, overloaded service entrance equipment, and improperly grounded meter enclosures — that create shock and fire risk to the customer premises occupants and utility field personnel who subsequently access the meter enclosure without awareness of the bypass condition. The adversarial perturbation required to suppress a tamper indicator classification in smart meter AI can be embedded in the visual field of the meter display photograph at a pixel density that is undetectable under standard field photography quality review, as AI-assisted revenue protection systems routinely process field photographs taken under varying lighting conditions, camera angles, and compression settings that create legitimate image quality variation masking adversarial perturbation patterns at human reviewer inspection resolution.

The regulatory and compliance consequences of adversarially suppressed smart meter anomaly detection span FERC Order 2222 DER compliance, state public utility commission tariff enforcement, and utility AML/fraud detection obligation dimensions. FERC Order 2222 (Participation of Distributed Energy Resources in Markets Operated by Regional Transmission Organizations and Independent System Operators) requires that DER aggregators participating in RTO/ISO markets meet metering accuracy and settlement verification requirements that depend on AMI meter data integrity; adversarial manipulation of Itron AI or Landis+Gyr AI meter display anomaly detection that suppresses tamper or bypass indicators for meters serving enrolled DER resources creates FERC Order 2222 settlement compliance failures at the RTO/ISO level, potentially affecting DER aggregation market participation authorisation for the affected aggregator. State public utility commission tariff enforcement obligations require utilities to investigate and recover revenues lost to documented metering fraud and energy theft; adversarial suppression of AI-detected metering anomalies that defers fraud investigation referral creates state PUC compliance exposure where the utility cannot demonstrate adequate revenue protection programme performance to state regulators conducting earnings review or rate case prudency assessments. Utility anti-money-laundering and fraud detection obligations — arising from state utility commission regulations, utility tariff provisions, and in some jurisdictions federal financial institution obligations where utilities offer financial products — require utilities to maintain audit trails of fraud detection tool performance; adversarial manipulation of AI-assisted meter anomaly detection tools that suppresses fraud indicators without generating audit records creates a compliance gap in the utility's fraud detection programme documentation. Threshold: 60 for smart meter display AI — elevated above substation and transmission thresholds to reflect the high-volume, high-frequency character of adversarial metering fraud and the compounding revenue exposure across large AMI populations.

3. Transmission line aerial inspection image injection (GE Vernova GridIQ AI, Siemens Energy Management AI)

Transmission line aerial inspection AI processes aerial survey photographs captured by drone inspection platforms, helicopter-borne survey camera systems, fixed-wing aerial patrol photography, and satellite-derived imagery submitted through AI-assisted transmission line condition assessment and maintenance planning tools that extract conductor sag condition measurements, insulator damage classifications, vegetation encroachment clearance distance determinations, tower structure corrosion severity grades, and hardware fitting condition ratings from aerial inspection image inputs, generating transmission line maintenance work order priorities, facility rating recalculation triggers, vegetation management intervention referrals, and emergency line de-energisation recommendations that govern whether transmission lines continue to operate at their rated facility thermal capacity or require derating, emergency reconductoring, or immediate outage for vegetation clearing or conductor replacement. GE Vernova GridIQ AI processes aerial transmission line inspection photographs and drone survey image streams through AI-assisted asset health management and transmission line condition monitoring tools at transmission owners and transmission operators globally, with AI-generated conductor sag assessments and vegetation encroachment clearance determinations directly informing NERC FAC-001-4 facility rating documentation and FAC-002-4 facility connection requirements compliance at bulk electric system transmission facilities. Siemens Energy Management AI processes aerial line inspection image sets and drone survey data through AI-assisted transmission asset management tools at transmission system operators worldwide, with AI condition grade outputs informing transmission planning study inputs, facility rating adjustments, and outage coordination scheduling decisions that affect interconnection-wide reliability assessments. The convergence of drone inspection platform adoption, AI-assisted image analysis, and NERC FAC-001-4 facility rating compliance obligations has created a high-volume, operationally critical aerial inspection image processing workflow at transmission owners and operators across the North American interconnections — a workflow that processes thousands of aerial inspection photographs per transmission line inspection cycle across tens of thousands of kilometres of bulk electric system transmission infrastructure.

The adversarial injection surface is the aerial conductor condition photograph, vegetation encroachment clearance image, insulator inspection photograph, and tower structure corrosion survey image submission pathway: transmission line aerial inspection images submitted through GE Vernova GridIQ AI or Siemens Energy Management AI transmission line condition assessment interfaces for AI conductor sag measurement, vegetation clearance determination, and structural deficiency classification. An adversarially crafted aerial transmission line inspection photograph — in which pixel perturbations applied to the conductor catenary curve measurement display, vegetation canopy height indicator, or insulator string damage evidence region on an aerial survey image cause the AI to classify the transmission span as meeting facility rating clearance and conductor condition requirements when the actual image documents a conductor sag exceedance requiring NERC FAC-001-4 facility thermal rating revision, a vegetation encroachment clearance violation triggering mandatory vegetation management intervention, or an insulator damage condition meeting the threshold for emergency line de-energisation — can suppress a condition classification that would otherwise mandate a facility rating reduction, a vegetation management work order, or an emergency line outage, allowing the transmission line to continue operating at its rated thermal capacity through a span whose actual physical condition creates conductor galloping risk, vegetation contact fault risk, and wildfire ignition risk from conductor-to-vegetation arc events in high fire threat districts. In Western North American transmission systems where aerial inspection AI is used to detect vegetation encroachment clearance violations on transmission lines traversing high fire threat districts, adversarial suppression of a vegetation clearance violation classification during an inspection cycle can allow a conductor-to-vegetation contact condition to persist through fire season months during which a single arc ignition event can initiate a wildfire with community-scale consequence, as demonstrated by the causal relationship between transmission line vegetation contact events and large wildfire events across California, Oregon, and Washington in recent years.

The regulatory and liability consequences of adversarially suppressed transmission line aerial inspection AI span NERC FAC-001-4 facility ratings, NERC FAC-002-4 facility interconnection requirements, FERC reliability standard enforcement, and wildfire liability dimensions of exceptional severity. NERC FAC-001-4 (Facility Ratings) requires transmission owners to establish and maintain facility ratings for their transmission facilities using a methodology that accounts for conductor thermal rating, structure clearance limitations, and vegetation management clearances; adversarial AI manipulation that suppresses a conductor sag exceedance or vegetation clearance violation that would otherwise require a facility rating revision creates a FAC-001-4 compliance failure in which the transmission owner's documented facility rating does not reflect the facility's actual operating limit, with NERC civil penalty exposure at the $1M+/day level. NERC FAC-002-4 (Facility Connection Requirements) requires transmission owners and transmission planners to verify that interconnected transmission facilities meet facility connection reliability requirements including clearance standards; adversarial suppression of an AI-detected clearance violation that affects FAC-002-4 compliance documentation creates interconnection reliability assessment failures at the transmission planning organisation level. FERC reliability standard enforcement under FERC Order 693 mandates compliance with all applicable NERC reliability standards including FAC-001 and FAC-002; FERC civil penalty proceedings for FAC-001 or FAC-002 violations arising from adversarially manipulated aerial inspection AI assessments assess penalty amounts based on the duration and severity of the compliance failure, the degree to which the violation created actual transmission reliability risk, and the corrective actions implemented. Beyond NERC/FERC enforcement, a transmission line wildfire ignition event caused by adversarially suppressed vegetation clearance AI detection creates inverse condemnation liability exposure under California Public Utilities Code § 2106 and similar state statutes in Western states, where transmission owners bear strict liability for wildfire damages caused by utility infrastructure operation regardless of negligence. Threshold: 50 for transmission line aerial inspection AI — the strictest threshold, reflecting wildfire ignition and grid reliability life-safety primacy.

4. SCADA and energy management system display injection (Schneider Electric EcoStruxure AI, OSIsoft PI System AI)

SCADA and energy management system display AI processes screenshots of distribution management system (DMS) operator workstation displays, energy management system (EMS) state estimator output displays, OSIsoft PI System trend visualisation screenshots, Schneider Electric EcoStruxure grid monitoring interface screenshots, and real-time power flow diagram display images submitted through AI-assisted grid anomaly detection, demand-supply imbalance identification, and fault location monitoring tools that extract grid anomaly alert classifications, demand-supply balance deviation measurements, fault location coordinate determinations, and contingency alarm priority rankings from EMS/SCADA display screenshot inputs, generating grid operator advisory notifications, automatic corrective action initiation triggers, and NERC EOP-004-4 reportable event identification flags that inform energy management centre operators, system operators, and reliability coordinators about real-time grid state deviations, contingency conditions, and event reporting obligations. Schneider Electric EcoStruxure AI is deployed across more than 200,000 electrical installations worldwide, including distribution utility operations centres, transmission operator energy management centres, industrial facility power management systems, and commercial building electrical infrastructure, processing EcoStruxure Grid monitoring display screenshots and distribution management system operator interface images through AI-assisted distribution automation, fault location isolation and service restoration (FLISR), and advanced distribution management system (ADMS) tools that govern automated switching operations, feeder reconfiguration decisions, and distribution transformer load management actions at distribution utility operators serving millions of customer endpoints. OSIsoft PI System AI is deployed at more than 22,000 customer sites including electricity generation facilities, transmission operators, distribution utilities, and large industrial customers, processing PI System SCADA trend display screenshots, event frame analysis interface images, and operational data historian visualisation screenshots through AI-assisted anomaly detection, performance monitoring, and event investigation tools that serve as the foundational operational data infrastructure layer for a substantial fraction of global electricity system monitoring — including generation dispatch monitoring, transmission constraint tracking, and distribution reliability performance analysis at utility operators whose combined coverage spans a significant portion of North American and European bulk electric system generation and transmission infrastructure. Eaton AI power quality monitoring and Enel X AI demand response management augment the SCADA/EMS display AI surface with power quality analyser display image processing and demand response event verification display screenshot analysis that generate power quality anomaly classifications and demand response curtailment verification determinations affecting FERC Order 2222 DER market participation compliance and state utility commission demand response programme performance reporting.

The adversarial injection surface is the EMS state estimator display screenshot, OSIsoft PI System trend visualisation image, Schneider Electric EcoStruxure distribution management display screenshot, and SCADA workstation interface photograph submission pathway: energy management system and SCADA display images submitted through Schneider Electric EcoStruxure AI or OSIsoft PI System AI monitoring interfaces for AI grid anomaly detection, demand-supply imbalance identification, and fault location marker extraction. An adversarially crafted EMS display screenshot — in which pixel perturbations applied to the contingency alarm indicator region, demand-supply balance deviation display, or fault location marker on an energy management centre workstation screenshot cause the AI to classify the grid state as within normal operating parameters when the actual display shows a contingency alarm condition, a demand-supply imbalance exceeding the balancing authority's operational tolerance, or a fault location marker indicating an active transmission or distribution system fault — can suppress a grid anomaly alert that would otherwise trigger grid operator corrective action, automatic SCADA remediation, or NERC EOP-004-4 event reporting, allowing a developing grid contingency condition to progress without AI-assisted operator advisory support during the time window in which early corrective action is most effective at preventing load shedding, cascading tripping, or interconnection-wide reliability events. In energy management centre environments where AI-assisted SCADA display monitoring tools provide secondary alert generation and anomaly detection support to grid operators managing complex grid state conditions across thousands of monitored assets, adversarial suppression of a contingency alarm classification in an EMS display screenshot creates an information gap in the AI advisory layer precisely at the moment when the grid state is most abnormal — the adversarial injection attack is most effective when the grid is under stress and AI advisory support is most operationally valuable to the grid operator. The physical scale of adversarial consequence in EMS/SCADA display AI injection is interconnection-level: a suppressed contingency alarm for a high-voltage transmission bus fault or a masked demand-supply imbalance indicator at a balancing authority boundary can, under stressed grid conditions, contribute to cascading contingency propagation across an interconnected transmission system serving tens of millions of customers.

The regulatory consequences of adversarially suppressed SCADA and EMS display AI monitoring span NERC CIP-007-6 electronic security perimeter requirements, NERC EOP-004-4 event reporting obligations, FERC Order 693 reliability standard enforcement, and state public utility commission outage reporting dimensions. NERC CIP-007-6 (Systems Security Management) requires responsible entities to implement ports and services management, security patch management, malicious code prevention, security event monitoring, and system access control for BES Cyber Systems within electronic security perimeters; AI-assisted SCADA and EMS display monitoring tools that process operational display screenshots from BES Cyber Systems constitute components of the electronic security perimeter monitoring architecture, and adversarial manipulation of these AI monitoring tools through crafted display screenshots creates a CIP-007-6 security event monitoring control failure with NERC civil penalty exposure at $1M+/day. NERC EOP-004-4 (Event Reporting) requires transmission operators, balancing authorities, reliability coordinators, generator operators, and distribution providers to report disturbances and unexpected occurrences that meet defined criteria within specified timeframes — including disturbances involving load loss exceeding specified thresholds, unexpected transmission facility outages, and automatic load shedding events; adversarial suppression of AI-generated EOP-004-4 event identification flags in SCADA monitoring tools that delays reportable event identification beyond the EOP-004-4 reporting timeline creates a reportable event timeliness compliance failure with NERC enforcement consequences. FERC Order 693 reliability standard enforcement covers both CIP-007-6 and EOP-004-4 violations arising from adversarially manipulated grid monitoring AI, with FERC civil penalty proceedings assessing penalty amounts that reflect the actual grid reliability consequences of the monitoring failure — including any load loss or generation dispatch impacts that occurred during the period when adversarial AI suppression prevented appropriate grid operator corrective action. Threshold: 55 for SCADA and EMS display AI — reflecting the electronic security perimeter and event reporting compliance dimensions while acknowledging that EMS display AI typically supplements primary SCADA alarm systems rather than serving as the sole contingency detection mechanism.

Integration: smart grid and power distribution AI image ingestion with Glyphward pre-scan

Smart grid and power distribution AI image ingestion flows from substation condition monitoring photograph APIs, AMI meter display photograph channels, aerial transmission line inspection data portals, and EMS/SCADA display screenshot interfaces into substation asset health AI, revenue protection AI, transmission line condition assessment AI, and grid event monitoring AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to substation maintenance work orders, energy theft investigation referrals, transmission facility rating records, or EMS event reports:

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"

# Smart grid & power distribution AI — NERC CIP-007-6, NERC CIP-014-3,
# FERC Order 693, FERC Order 2222, NERC FAC-001-4, NERC FAC-002-4,
# NERC EOP-004-4, state PUC. Suppression of substation thermal degradation,
# transmission line conductor sag, EMS grid fault indicators, and smart meter
# tamper classifications create grid reliability, wildfire, and life-safety risk.
THRESHOLD_LIFE_SAFETY   = 50  # substation condition, transmission line (strictest)
THRESHOLD_SCADA_EMS     = 55  # SCADA/EMS display (CIP-007-6, EOP-004-4)
THRESHOLD_SMART_METER   = 60  # smart meter display (revenue protection, FERC 2222)


class GridAIContext(str, Enum):
    SUBSTATION_CONDITION = "substation_condition"  # GE Vernova GridIQ, Siemens, ABB Ability
    SMART_METER_DISPLAY  = "smart_meter_display"   # Itron AI, Landis+Gyr AI
    TRANSMISSION_LINE    = "transmission_line"      # GE Vernova GridIQ, Siemens Energy Mgmt
    SCADA_DISPLAY        = "scada_display"          # Schneider EcoStruxure, OSIsoft PI System


def threshold_for(context: GridAIContext) -> int:
    if context == GridAIContext.SMART_METER_DISPLAY:
        return THRESHOLD_SMART_METER
    if context == GridAIContext.SCADA_DISPLAY:
        return THRESHOLD_SCADA_EMS
    return THRESHOLD_LIFE_SAFETY


async def scan_grid_ai_image(
    image_path: str | Path,
    context: GridAIContext,
    utility_id_hash: str,    # SHA-256 of utility operator / transmission owner ID
    asset_ref: str,          # e.g. "XFMR-BERK-T3", "METER-0042881", "LINE-230KV-A7"
    inspection_run_id: str,  # e.g. drone survey run ID, AMI inspection session ID
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a smart grid or power distribution AI image for adversarial injection
    payloads before forwarding to substation condition, smart meter revenue
    protection, transmission line inspection, or SCADA/EMS monitoring AI.

    Raises AdversarialGridAIImageError if score meets threshold:
      - SUBSTATION_CONDITION: threshold 50; NERC CIP-014-3; FERC Order 693;
                              $1M+/day violation; life-safety/grid stability
      - TRANSMISSION_LINE:    threshold 50; NERC FAC-001-4; FAC-002-4;
                              FERC reliability standard; wildfire consequence
      - SCADA_DISPLAY:        threshold 55; NERC CIP-007-6; NERC EOP-004-4;
                              FERC Order 693; electronic security perimeter
      - SMART_METER_DISPLAY:  threshold 60; FERC Order 2222; state PUC tariff;
                              utility AML/fraud detection obligations
    """
    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": {
                "grid_context":       context.value,
                "utility_id_hash":    utility_id_hash,
                "asset_ref":          asset_ref,
                "inspection_run_id":  inspection_run_id,
                "client_scan_id":     client_scan_id,
                "image_sha256":       image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "utility_id_hash":    utility_id_hash,
        "asset_ref":          asset_ref,
        "inspection_run_id":  inspection_run_id,
        "grid_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_grid_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialGridAIImageError(
            f"Grid AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"utility={utility_id_hash} asset={asset_ref}"
        )
    return result


async def write_grid_audit_record(record: dict) -> None:
    """Persist audit record to utility NERC CIP compliance audit store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialGridAIImageError(Exception):
    """Raised when a smart grid or power distribution AI image exceeds the adversarial injection threshold."""
    pass

Call scan_grid_ai_image() with GridAIContext.SUBSTATION_CONDITION before forwarding transformer thermal inspection photographs and switchgear condition images to GE Vernova GridIQ AI, Siemens Energy Management AI, or ABB Ability AI substation monitoring tools — the highest life-safety integration point, where adversarial suppression of a thermal degradation flag defers required transformer de-energisation and allows thermal runaway progression in a high-voltage transmission substation asset. Call with GridAIContext.TRANSMISSION_LINE for GE Vernova or Siemens Energy Management AI aerial inspection images before AI conductor sag measurement and vegetation clearance classification, preserving image_sha256 as the forensic anchor for NERC FAC-001-4 facility rating documentation audit and wildfire liability evidence chain. Call with GridAIContext.SCADA_DISPLAY for Schneider Electric EcoStruxure AI and OSIsoft PI System AI EMS display screenshots before AI grid anomaly and fault location classification, with asset_ref encoding the EMS substation or feeder identifier for NERC CIP-007-6 security event monitoring audit trail and EOP-004-4 event reporting timeline documentation. Call with GridAIContext.SMART_METER_DISPLAY for Itron AI or Landis+Gyr AI meter display photographs before AI tamper indicator detection and abnormal consumption classification, with inspection_run_id linking the Glyphward scan record to the specific AMI field inspection session or revenue protection programme audit cycle for state PUC tariff enforcement and FERC Order 2222 DER settlement verification documentation. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Indirect prompt injection via image — foundational attack pattern underlying all four smart grid and power distribution AI injection surfaces; covers how adversarial pixel-level perturbations cause AI misclassification without detectable visual artefacts at human review resolution, directly applicable to substation thermal photograph manipulation and EMS display screenshot injection.
• Vision-language model security — technical architecture of adversarial image attacks against vision-language models including pixel perturbation classes applicable to substation condition monitoring AI and aerial transmission line inspection image injection against GE Vernova GridIQ AI and Siemens Energy Management AI.
• Manufacturing quality inspection AI prompt injection — industrial AI injection context covering the broader class of physical inspection photograph manipulation applicable to substation equipment condition assessment injection and power quality monitoring AI adversarial image attacks.
• Water treatment and environmental AI prompt injection — adjacent critical infrastructure AI injection context covering adversarial image attacks against utility operational AI platforms with overlapping NERC/FERC and state utility regulatory frameworks applicable to grid operations and environmental compliance monitoring.
• Free tier — 10 scans/day, no card required — start scanning smart grid and power distribution AI images at development volumes before committing to a production plan.