Prompt injection in architecture, engineering and surveying AI

Four adversarial injection surfaces in architecture, engineering and surveying AI

1. Structural engineering drawing injection (Autodesk Revit AI, Ansys SimAI)

Structural engineering drawing AI processes PE-stamped structural drawing scan images, BIM model structural analysis visualisations, and finite element analysis result display graphics from Autodesk Revit AI at more than 2 million architecture and engineering users globally, Ansys SimAI at engineering firms conducting AI-assisted structural performance simulation and finite element analysis, Bentley Systems OpenInfrastructure AI at infrastructure engineering organisations conducting bridge and civil infrastructure structural analysis, PTC Creo Generative Design AI at manufacturing and industrial engineering organisations, and integrated BIM platform structural analysis AI tools, extracting structural engineering compliance classifications — load path adequacy assessments, structural member capacity utilisation ratios, seismic demand-to-capacity compliance indicators, gravity and lateral load path continuity verifications, connection detail compliance flags, and IBC structural safety code compliance determinations — from PE-stamped structural drawing scan image inputs in AI-assisted structural engineering analysis pipelines, generating structural compliance verification records, deficiency identification reports, code compliance determination outputs, and structural peer review documentation entries that licensed structural engineers and building official plan reviewers depend upon for AIA A201 Article 3.7.4 contractor submittal review, ASCE 7-22 structural load standard compliance verification, and IBC Chapter 16 structural design load requirements compliance. Autodesk Revit AI’s structural analysis tools process BIM model structural system configurations through AI-assisted structural load path analysis and code compliance checking tools that licensed structural engineers use to verify whether proposed structural systems meet ASCE 7-22 seismic, wind, and gravity load requirements and IBC Chapter 16 structural load combination requirements before PE stamp application. Ansys SimAI’s AI-assisted finite element analysis tools process structural performance simulation result visualisations through AI-assisted structural performance compliance and safety factor verification tools at engineering design firms where AI-generated structural analysis reports inform the PE stamp application decision for client-submitted structural engineering designs.

The adversarial injection surface is the PE-stamped structural drawing scan and structural analysis visualisation submission pathway: Autodesk Revit AI or Ansys SimAI PE-stamped structural drawing scan images and structural analysis result visualisations submitted through AI-assisted load path compliance verification, structural member capacity assessment, and IBC code compliance checking tools for AI structural deficiency identification and PE-stamped design acceptability determination. An adversarially crafted Autodesk Revit AI structural drawing scan image — in which pixel perturbations applied to the lateral load path connection detail region, the seismic demand-to-capacity ratio display marker, or the gravity load path continuity indicator in a PE-stamped structural drawing scan cause the AI to classify a structural system with a load path deficiency creating IBC Chapter 16 code non-compliance as a fully compliant structural design meeting ASCE 7-22 and IBC structural requirements when the actual drawing documents a structural connection or member capacity deficiency requiring design correction before PE stamp application — can suppress a structural deficiency detection that would otherwise generate a code compliance deficiency report, a design correction requirement, and a structural peer review documentation entry. In high-volume engineering design review environments where Autodesk Revit AI or Ansys SimAI processes thousands of structural drawing submissions per project design phase without individual licensed structural engineer review of every AI compliance classification, adversarial suppression of structural load path deficiency detections allows non-compliant structural designs to be PE-stamped and submitted for building department plan review with IBC structural safety and professional engineer malpractice liability consequences.

The professional liability and structural safety consequences of adversarially suppressed structural load path deficiency detection in structural engineering drawing AI span PE stamp professional malpractice, ASCE 7-22, IBC structural, AIA B101 professional liability, and building failure public safety dimensions. Licensed professional engineers bear personal professional liability for the structural designs they PE-stamp; state engineering licensing boards establish PE professional responsibility standards requiring that licensed engineers personally verify that PE-stamped structural designs comply with applicable engineering standards and building codes. Adversarial manipulation of Autodesk Revit AI or Ansys SimAI structural analysis tools that suppresses IBC structural load path deficiency detection in a structural system that the PE stamps in reliance on AI-assisted compliance verification creates PE stamp professional malpractice liability when the structurally deficient design causes a building failure or requires costly post-construction remediation — because the PE is professionally responsible for the accuracy of the AI-assisted structural compliance analysis on which the PE stamp application is based. ASCE 7-22 (Minimum Design Loads and Associated Criteria for Buildings and Other Structures) specifies minimum structural load standards adopted by IBC as the reference structural load standard for building code compliance; adversarial suppression of structural AI detection of ASCE 7-22 load combination deficiencies creates IBC Chapter 16 structural safety non-conformance with building failure, public safety, and structural engineering professional liability dimensions. AIA B101 Standard Form of Agreement between Owner and Architect specifies that the architect’s professional services include structural engineering coordination and code compliance verification; adversarial manipulation of architect-deployed structural AI tools that suppresses structural deficiency detection creates AIA B101 professional service performance obligation exposure. Threshold: 65 for structural engineering drawing AI — reflecting the PE stamp professional malpractice, ASCE 7-22, IBC structural safety, and building failure public safety dimensions of suppressed structural load path deficiency detection.

2. Geospatial survey data visualisation injection (ESRI ArcGIS AI, Topcon AI)

Geospatial survey data visualisation AI processes survey boundary data display images, parcel boundary identification graphics, elevation contour mapping visualisations, and ALTA/NSPS land survey plat display images from ESRI ArcGIS AI at more than 300,000 organisations globally including government land agencies, title insurance companies, and engineering survey firms, Topcon AI precision positioning and survey data analysis tools at professional survey operations, Trimble AI survey data management and geospatial analysis tools at land survey and engineering firms, and integrated land survey platform AI tools, extracting survey accuracy and boundary compliance classifications — parcel boundary line accuracy assessments, easement encroachment detections, survey monument proximity indicators, elevation datum compliance verifications, and ALTA/NSPS minimum accuracy standard conformance determinations — from survey boundary visualisation image inputs in AI-assisted land survey quality assurance and boundary dispute analysis pipelines, generating survey accuracy certification records, encroachment notification documentation, boundary dispute expert witness report inputs, and title insurance underwriting documentation entries that licensed land surveyors and title insurance underwriters depend upon for ALTA/NSPS minimum standard detail requirements compliance and real estate transaction title report accuracy. ESRI ArcGIS AI’s survey boundary analysis tools process parcel boundary and easement data visualisations through AI-assisted encroachment detection and boundary accuracy verification tools that licensed land surveyors use to verify ALTA/NSPS survey accuracy requirements before survey plat certification and professional land surveyor stamp application. Topcon AI’s precision positioning data analysis tools process survey measurement result visualisations through AI-assisted boundary accuracy and ALTA/NSPS minimum accuracy standard verification tools at professional survey operations where AI-generated boundary accuracy certifications inform the licensed land surveyor’s ALTA/NSPS survey plat stamp decision.

The adversarial injection surface is the survey boundary data visualisation and ALTA/NSPS survey plat display image submission pathway: ESRI ArcGIS AI or Topcon AI survey boundary visualisation images and ALTA/NSPS land survey plat display graphics submitted through AI-assisted boundary encroachment detection, easement mapping accuracy verification, and ALTA/NSPS minimum standard conformance checking tools for AI boundary compliance identification and licensed land surveyor certification determination. An adversarially crafted ESRI ArcGIS AI parcel boundary visualisation image — in which pixel perturbations applied to the property line boundary marker display, the easement boundary encroachment indicator region, or the survey monument proximity reference in a parcel boundary data visualisation cause the AI to classify a survey boundary line that encroaches on an adjoining property or impermissibly violates an easement boundary as a fully compliant boundary configuration meeting ALTA/NSPS minimum accuracy standard requirements when the actual survey data documents an encroachment creating a boundary dispute liability — can suppress an encroachment detection flag that would otherwise generate a boundary dispute documentation entry, a title insurance underwriter notification, and a survey accuracy certification deficiency record. In large-scale land survey and real estate transaction environments where ESRI ArcGIS AI or Topcon AI processes survey boundary visualisations across hundreds of property boundary determinations per month without individual licensed surveyor review of every AI encroachment classification, adversarial suppression of boundary encroachment detections allows encroachments to be certified as ALTA/NSPS compliant with real estate title insurance, boundary dispute litigation, and professional land surveyor malpractice consequences.

The professional liability and real estate transaction consequences of adversarially suppressed boundary encroachment detection in geospatial survey AI span ALTA/NSPS minimum standard requirements, licensed land surveyor professional malpractice, real estate title insurance underwriting, and real property boundary dispute litigation dimensions. ALTA/NSPS Minimum Standard Detail Requirements for ALTA/NSPS Land Title Surveys (2021) specify the accuracy standards and survey content requirements that licensed land surveyors must meet for surveys accepted by title insurance companies in real estate purchase and mortgage transactions; adversarial manipulation of ESRI ArcGIS AI boundary analysis that suppresses encroachment detection in an ALTA/NSPS survey creates survey accuracy non-conformance that, if discovered in post-transaction boundary dispute proceedings, creates licensed land surveyor professional malpractice liability for the encroachment that the adversarially clean AI survey failed to identify. Title insurance companies rely on ALTA/NSPS survey certifications as a primary due diligence input for underwriting the title insurance policies that insure real estate buyers and mortgage lenders against boundary encroachment and survey accuracy claims; adversarially clean ESRI ArcGIS AI survey certification that fails to document an actual boundary encroachment creates title insurance claim exposure when the undisclosed encroachment is subsequently identified in boundary dispute proceedings initiated by an adjoining property owner. Real property boundary dispute litigation in state courts depends on licensed land surveyor expert witness testimony supported by survey data visualisation evidence — ESRI ArcGIS AI-generated boundary analysis records and visualisation outputs that adversarially suppress encroachment indicators create challenges for the adversely affected property owner’s boundary dispute litigation evidence base when AI-generated records document a clear boundary condition that the actual survey data does not support. Threshold: 65 for geospatial survey data visualisation AI — reflecting the ALTA/NSPS survey accuracy, licensed surveyor professional malpractice, title insurance underwriting, and boundary dispute litigation dimensions of suppressed boundary encroachment detection.

3. Environmental site assessment photograph injection (WSP/AECOM digital delivery AI)

Environmental site assessment photograph AI processes Phase I Environmental Site Assessment site investigation photograph documentation images, recognised environmental condition evidence photographs, historical land use documentation images, and Phase II subsurface investigation field photograph records from WSP/AECOM digital delivery AI at major environmental engineering and consulting firms conducting ASTM E 1527-21 Phase I ESA assessments for commercial real estate transactions, Hexagon Smart Construction AI at construction site investigation programme operations, and integrated environmental assessment platform AI document analysis tools, extracting Phase I ESA recognised environmental condition REC classification assessments — UST presence indicator detections, petroleum hydrocarbon staining visual evidence identifications, recognised environmental condition evidence photograph accuracy verifications, historical use documentation completeness flags, and ASTM E 1527-21 all appropriate inquiry AAI compliance determinations — from Phase I ESA site investigation photograph inputs in AI-assisted environmental due diligence analysis pipelines, generating REC classification records, Phase I ESA report accuracy certifications, all appropriate inquiry compliance determination records, and environmental transaction screen documentation entries that environmental professionals and commercial real estate transaction parties depend upon for CERCLA §101(35)(B) innocent landowner defence qualification and commercial real estate transaction environmental liability due diligence compliance. WSP and AECOM Phase I ESA digital delivery AI processes environmental site investigation photograph documentation through AI-assisted REC identification and ASTM E 1527-21 AAI compliance verification tools that environmental professionals use to verify that Phase I ESA site investigation photograph documentation meets ASTM E 1527-21 recognised environmental condition documentation standards before Phase I ESA report certification and professional environmental assessor signature. CERCLA §101(35)(B) innocent landowner defence requires that a party claiming innocent landowner status demonstrate that it conducted all appropriate inquiry into the previous ownership and uses of the facility before acquiring the property — Phase I ESA documentation that meets ASTM E 1527-21 standards is the primary evidence of AAI compliance for CERCLA innocent landowner defence qualification.

The adversarial injection surface is the Phase I ESA site investigation photograph and environmental documentation image submission pathway: WSP/AECOM or Hexagon AI Phase I ESA site photograph documentation images submitted through AI-assisted recognised environmental condition identification, ASTM E 1527-21 documentation completeness verification, and CERCLA AAI compliance determination tools for AI REC classification and Phase I ESA report accuracy certification. An adversarially crafted Phase I ESA site investigation photograph — in which pixel perturbations applied to the petroleum hydrocarbon staining visual evidence region, the UST vent pipe or fill port visual indicator, or the recognised environmental condition evidence marker in a site investigation photograph cause the AI to classify a site investigation photograph documenting a recognised environmental condition meeting ASTM E 1527-21 REC identification criteria as a standard no-REC site condition when the actual photograph documents soil staining, UST infrastructure, or environmental condition evidence requiring Phase I ESA REC documentation — can suppress a REC identification that would otherwise generate a Phase I ESA recognised environmental condition classification, a Phase I report REC section documentation entry, and a CERCLA AAI compliance qualification notation. In high-volume commercial real estate transaction environments where environmental consulting firms use AI-assisted Phase I ESA photograph analysis tools to process documentation for multiple concurrent site assessments without individual environmental professional review of every photograph AI classification, adversarial suppression of REC identifications allows Phase I ESA reports to certify CERCLA AAI compliance without documenting environmental conditions that ASTM E 1527-21 requires to be documented as RECs.

The CERCLA liability and professional consequences of adversarially suppressed REC identification in Phase I ESA photograph AI span CERCLA §101(35)(B) innocent landowner defence, ASTM E 1527-21 all appropriate inquiry compliance, environmental professional malpractice, and commercial real estate transaction due diligence liability dimensions. CERCLA §107(b)(3) innocent landowner defence exemption requires that the innocent landowner demonstrate that the contamination was caused solely by a third party and that the owner exercised due care with respect to the hazardous substance and took precautions against foreseeable acts of such third parties; CERCLA §101(35)(B) specifies that all appropriate inquiry compliance under the AAI rule (40 CFR Part 312 implementing ASTM E 1527-21) is necessary for qualifying for the bona fide prospective purchaser exemption. Adversarial suppression of Phase I ESA photograph AI REC identifications that produces a Phase I ESA report without required REC documentation creates CERCLA AAI compliance failures when the missing REC documentation undermines the property purchaser’s CERCLA innocent landowner defence in subsequent EPA or state environmental enforcement proceedings. Environmental professional malpractice liability attaches to licensed environmental assessors who certify Phase I ESA reports that fail to identify RECs meeting ASTM E 1527-21 identification criteria; adversarial manipulation of WSP/AECOM AI phase I ESA photograph analysis that suppresses REC identifications creates environmental professional malpractice exposure for the certifying environmental assessor when the undisclosed REC causes CERCLA liability for the property purchaser who relied on the adversarially incomplete Phase I report. Commercial real estate transaction title insurance and representation and warranty insurance policies in M&A transactions involving contaminated property depend on Phase I ESA REC documentation for environmental liability coverage underwriting; adversarially suppressed Phase I ESA AI REC documentation creates title insurance and RWI coverage gaps with post-closing environmental discovery liability dimensions. Threshold: 60 for environmental site assessment photograph AI — reflecting the CERCLA innocent landowner defence, ASTM E 1527-21 AAI compliance, environmental professional malpractice, and commercial real estate transaction due diligence dimensions of suppressed REC identification.

4. As-built record scan injection (Hexagon Smart Construction AI, Trimble AI)

As-built record scan AI processes construction as-built record document scan images, structural shop drawing submittal photographs, MEP installation verification photographs, and building commissioning record document images from Hexagon Smart Construction AI at major construction programme management operations, Trimble MEP AI at MEP engineering and construction management firms, Autodesk BIM 360 AI construction document management tools at architecture and construction firms, and integrated construction document management platform AI tools, extracting as-built accuracy and construction compliance classifications — structural installation compliance with permit-approved drawings indicators, MEP system installation accuracy verifications, building commissioning documentation completeness flags, and AIA G704 certificate of substantial completion accuracy determinations — from as-built record document scan image inputs in AI-assisted construction quality assurance and project closeout management pipelines, generating construction completion certification records, AIA G702 payment application documentation accuracy entries, certificate of occupancy compliance determinations, and construction contract closeout documentation records that owners, contractors, and design professionals depend upon for AIA A201 Article 3.5 contractor warranty obligation documentation and building official certificate of occupancy issuance. Hexagon Smart Construction AI processes construction quality documentation photographs and as-built record scans through AI-assisted compliance verification tools at major construction programme operations including commercial, healthcare, and infrastructure construction projects where AI-generated as-built accuracy certifications inform the AIA G704 certificate of substantial completion issuance and building official certificate of occupancy determination. Trimble MEP AI’s construction document management tools process MEP installation verification photographs and as-built record images through AI-assisted installation accuracy and MEP code compliance verification tools at mechanical, electrical and plumbing engineering and construction organisations where AI-assisted as-built accuracy verification generates the construction closeout documentation records owners use for facility management and building commissioning compliance.

The adversarial injection surface is the as-built construction record document scan and installation verification photograph submission pathway: Hexagon Smart Construction AI or Trimble AI as-built record scans and MEP installation photographs submitted through AI-assisted construction compliance verification, AIA G704 substantial completion accuracy assessment, and certificate of occupancy eligibility determination tools for AI construction completion accuracy classification and building official compliance documentation. An adversarially crafted Hexagon Smart Construction AI as-built record scan image — in which pixel perturbations applied to the structural installation deviation indicator, the MEP system installation specification non-conformance marker, or the building commissioning deficiency visual in an as-built record scan image cause the AI to classify a construction completion record with material deviations from permit-approved drawings as a fully compliant substantial completion record meeting AIA G704 and certificate of occupancy requirements when the actual as-built scan documents material construction deficiencies requiring remediation before substantial completion certification — can suppress a construction deficiency detection that would otherwise generate an AIA A201 contractor warranty deficiency notice, a substantial completion certification hold, and a certificate of occupancy eligibility determination deficiency record. In large construction programme management environments where Hexagon Smart Construction AI or Trimble AI processes thousands of as-built record scans across multiple construction programme phases without individual design professional review of every AI compliance classification, adversarial suppression of material construction deficiency detections allows non-conforming construction to be certified as substantially complete with AIA A201 contractor warranty, building official certificate of occupancy, and construction defect liability consequences.

The construction defect and professional liability consequences of adversarially suppressed construction deficiency detection in as-built record scan AI span AIA A201 contractor warranty, building official certificate of occupancy, architect professional standard of care, and construction defect litigation dimensions. AIA A201 Article 3.5 (Warranty) requires contractors to warrant that the work meets the requirements of the contract documents; adversarial suppression of Hexagon Smart Construction AI or Trimble AI as-built compliance verification that generates clean completion records for construction work with material deviations from permit-approved drawings creates AIA A201 warranty obligation enforcement challenges when the contractor’s adversarially clean as-built documentation is used to support claims that the non-conforming construction meets contract requirements. Building official certificate of occupancy issuance depends on documentation demonstrating that construction complies with the permit-approved drawings and applicable building code requirements; adversarially clean AI as-built record scan certifications that suppress material construction deficiency detections create certificate of occupancy issuance irregularity exposure when the certified non-conforming construction is subsequently identified in building code enforcement inspections or construction defect expert investigations. Architect professional standard of care under AIA B101 requires that the architect’s construction administration services include reviewing contractor submittals and as-built records for general conformance with the design intent; adversarial manipulation of architect-deployed AI as-built analysis tools that suppresses construction deficiency detection creates AIA B101 construction administration standard of care exposure for the architect whose AI-generated compliance verification records failed to identify material construction deviations. Threshold: 60 for as-built record scan AI — reflecting the AIA A201 warranty, certificate of occupancy, architect professional standard of care, and construction defect litigation dimensions of suppressed construction deficiency detection.

Integration: architecture, engineering and surveying AI image ingestion with Glyphward pre-scan

Architecture, engineering and surveying AI image ingestion flows from Autodesk Revit and Ansys SimAI PE-stamped structural drawing scan APIs, ESRI ArcGIS and Topcon survey boundary visualisation image channels, WSP/AECOM and Hexagon Phase I ESA site photograph interfaces, and Hexagon Smart Construction and Trimble MEP as-built record scan platforms into structural load path compliance and IBC code checking AI, boundary encroachment detection and ALTA/NSPS certification AI, Phase I ESA REC identification and CERCLA AAI compliance AI, and as-built accuracy and certificate of occupancy eligibility AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to PE stamp structural compliance records, ALTA/NSPS survey certifications, Phase I ESA REC documentation, or as-built construction completion 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"

# Architecture, engineering & surveying AI — PE stamp professional malpractice;
# ASCE 7-22 structural load standard; IBC Chapter 16; ALTA/NSPS minimum
# standard detail requirements; ASTM E 1527-21 Phase I ESA; CERCLA §101(35)(B)
# innocent landowner defence; AIA A201 contractor warranty.
THRESHOLD_STRUCTURAL_AI      = 65  # Autodesk/Ansys; PE stamp; ASCE 7; IBC
THRESHOLD_SURVEY_AI          = 65  # ArcGIS/Topcon; ALTA/NSPS; title insurance
THRESHOLD_PHASE1_ESA_AI      = 60  # WSP/AECOM; ASTM E 1527-21; CERCLA AAI
THRESHOLD_AS_BUILT_AI        = 60  # Hexagon/Trimble; AIA A201; cert of occupancy


class AESSurveyingAIContext(str, Enum):
    STRUCTURAL_AI  = "structural_ai"   # Autodesk Revit, Ansys — load path/IBC
    SURVEY_AI      = "survey_ai"       # ArcGIS, Topcon — ALTA/NSPS boundary
    PHASE1_ESA_AI  = "phase1_esa_ai"  # WSP/AECOM — ASTM E 1527-21 REC
    AS_BUILT_AI    = "as_built_ai"    # Hexagon, Trimble — AIA A201/cert occupancy


def threshold_for(context: AESSurveyingAIContext) -> int:
    mapping = {
        AESSurveyingAIContext.STRUCTURAL_AI:  THRESHOLD_STRUCTURAL_AI,
        AESSurveyingAIContext.SURVEY_AI:      THRESHOLD_SURVEY_AI,
        AESSurveyingAIContext.PHASE1_ESA_AI:  THRESHOLD_PHASE1_ESA_AI,
        AESSurveyingAIContext.AS_BUILT_AI:    THRESHOLD_AS_BUILT_AI,
    }
    return mapping[context]


async def scan_aes_ai_image(
    image_path: str | Path,
    context: AESSurveyingAIContext,
    firm_id_hash: str,           # SHA-256 of AE firm or survey organisation ID
    project_submission_ref: str, # e.g. "RVT-PROJ-2026-44921", "ARC-SURV-88841"
    document_session_id: str,    # structural drawing session, survey plat run, ESA session
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan an AES AI image for adversarial injection payloads before forwarding
    to structural load path compliance, survey boundary encroachment detection,
    Phase I ESA REC identification, or as-built construction accuracy AI systems.

    Raises AdversarialAESAIImageError if score meets threshold:
      - STRUCTURAL_AI:  threshold 65; PE malpractice; ASCE 7-22; IBC structural
      - SURVEY_AI:      threshold 65; ALTA/NSPS; title insurance; boundary dispute
      - PHASE1_ESA_AI:  threshold 60; ASTM E 1527-21; CERCLA §101(35)(B) AAI
      - AS_BUILT_AI:    threshold 60; AIA A201 warranty; cert of occupancy
    """
    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": {
                "aes_context":             context.value,
                "firm_id_hash":            firm_id_hash,
                "project_submission_ref":  project_submission_ref,
                "document_session_id":     document_session_id,
                "client_scan_id":          client_scan_id,
                "image_sha256":            image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "firm_id_hash":            firm_id_hash,
        "project_submission_ref":  project_submission_ref,
        "document_session_id":     document_session_id,
        "aes_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_aes_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialAESAIImageError(
            f"AES AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"firm={firm_id_hash} ref={project_submission_ref}"
        )
    return result


async def write_aes_audit_record(record: dict) -> None:
    """Persist audit record to AES professional liability and compliance documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialAESAIImageError(Exception):
    """Raised when an architecture, engineering or surveying AI image exceeds the adversarial injection threshold."""
    pass

Call scan_aes_ai_image() with AESSurveyingAIContext.STRUCTURAL_AI before forwarding Autodesk Revit AI or Ansys SimAI PE-stamped structural drawing scans to structural load path compliance and IBC code checking AI — with project_submission_ref linking the Glyphward scan to the specific project structural submission for PE stamp professional liability and ASCE 7-22 compliance audit trail documentation. Call with AESSurveyingAIContext.SURVEY_AI for ESRI ArcGIS or Topcon survey boundary visualisation images before AI boundary encroachment detection and ALTA/NSPS accuracy certification, with document_session_id as the survey plat run identifier for title insurance underwriting and boundary dispute expert witness documentation. Call with AESSurveyingAIContext.PHASE1_ESA_AI for WSP/AECOM Phase I ESA site investigation photographs before AI REC identification and ASTM E 1527-21 AAI compliance determination, with project_submission_ref for CERCLA innocent landowner defence qualification audit trail. Call with AESSurveyingAIContext.AS_BUILT_AI for Hexagon Smart Construction or Trimble MEP as-built record scans before AI construction compliance and AIA G704 substantial completion accuracy verification, with document_session_id for certificate of occupancy and AIA A201 warranty audit documentation. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Indirect prompt injection via image — foundational attack pattern underlying all four AES AI injection surfaces; covers how adversarial pixel-level perturbations cause AI misclassification without detectable visual artefacts at human review resolution.
• Construction and BIM AI prompt injection — related construction AI injection context covering OSHA, structural safety, and BIM platform adversarial image attack surfaces applicable to as-built record and structural drawing contexts.
• Vision-language model security — technical architecture of adversarial image attacks applicable to structural drawing scan AI, geospatial visualisation AI, and Phase I ESA photograph AI processing pipelines.
• Free tier — 10 scans/day, no card required — start scanning architecture, engineering and surveying AI document images at development volumes before committing to a production plan.