Prompt injection in construction and BIM AI

Four adversarial injection surfaces in construction and BIM AI

1. Construction jobsite safety camera photograph injection (Procore AI, Smartvid.io AI, Doxel AI)

Construction jobsite safety camera photograph AI processes images captured by fixed jobsite safety cameras, superintendent and safety manager smartphone cameras, subcontractor foreman inspection photographs, and third-party safety observation photographs submitted through Procore AI safety management integrations, Smartvid.io AI safety analytics platforms, and Doxel AI construction monitoring tools that extract PPE compliance indicators — hard hat, high-visibility vest, safety glasses, fall protection harness status — fall protection deficiency flags, housekeeping hazard identifications, scaffolding compliance indicators, and OSHA reportable near-miss pattern scores from jobsite photograph inputs, generating safety observation records, corrective action notifications, safety performance KPI scores, and OSHA-reportable incident probability assessments that OSHA-covered construction employers depend upon for subpart C-X compliance program management, subcontractor safety accountability documentation, and OSHA 300 log recordkeeping with 29 CFR Part 1926 regulatory obligations. Smartvid.io AI has processed more than ten million safety observations from construction jobsite photographs, with AI-assisted safety classification tools deployed at Skanska USA, Turner Construction, Whiting-Turner, and other ENR Top 400 general contractors as the primary automated safety observation scoring platform; its AI outputs directly inform safety program corrective action priorities, subcontractor safety pre-qualification evaluation, and OSHA compliance audit documentation that GC safety directors rely upon for program effectiveness measurement and OSHA inspection preparation. Procore AI integrates safety photograph analysis into its project management platform deployed across more than two million active construction projects, with AI-assisted safety observation classification tools processing superintendent and foreman photograph submissions and generating safety compliance records that link to project punch lists, subcontractor performance evaluations, and AIA payment application safety compliance certifications in GC-to-owner project reporting workflows.

The adversarial injection surface is the jobsite safety camera photograph submission pathway: safety observation photographs submitted through Procore AI safety integrations or Smartvid.io AI safety analytics platforms for AI PPE compliance classification, fall protection deficiency scoring, and OSHA near-miss pattern detection. An adversarially crafted jobsite safety photograph — in which pixel perturbations applied to the worker hard hat absence region, fall protection harness connection point visual indicator, or scaffolding guardrail gap marker on a jobsite safety camera image cause the AI to classify a worker not wearing required PPE, or a scaffold with an OSHA-reportable fall protection gap, as compliant standard safe working conditions when the actual image documents a 29 CFR Part 1926 subpart E (PPE) or subpart Q (concrete and masonry) violation — can suppress an OSHA compliance deficiency identification that would otherwise generate a corrective action notification and safety program documentation record. In high-volume construction safety monitoring environments where Smartvid.io AI processes thousands of jobsite safety photographs per day across active project sites without individual human safety manager review of each AI-scored observation, adversarial suppression of PPE non-compliance classifications across a project cohort allows workers to remain in OSHA-violating conditions without generating the corrective action record that the GC’s safety program requires for OSHA 300 log accuracy, subcontractor safety accountability, and OSHA inspection preparation purposes.

The regulatory consequences of adversarially suppressed PPE violation and fall protection deficiency detection in construction jobsite safety camera AI span OSHA 29 CFR Part 1926 willful citation, workers’ compensation, and contractor liability dimensions of exceptional severity. OSHA 29 CFR Part 1926 subpart C establishes general safety and health provisions for construction employers; OSHA §5(a)(1) (the General Duty Clause) requires employers to provide workplaces free from recognised hazards likely to cause death or serious physical harm; adversarial manipulation of AI safety observation tools that suppresses recognition of fall protection gaps or PPE non-compliance converts a recognisable hazard into an unrecorded one, creating a General Duty Clause violation exposure that OSHA inspectors treat as willful when employer AI safety tools were technically capable of detecting the condition. OSHA willful citations carry civil penalties up to $156,259 per violation as of 2026 OSHA penalty adjustment; repeat willful citations double the maximum; and OSHA willful citations in cases where a worker suffers a fatality create misdemeanor criminal liability under 29 USC §666(e) with potential imprisonment. Construction employer workers’ compensation experience modification rates incorporate recordable injury frequency; adversarial suppression of safety observation corrective actions that prevents hazard elimination increases the probability of recordable injury events that elevate the employer’s experience modification rate and workers’ compensation premium for three policy years following the incident date. Threshold: 55 for construction jobsite safety camera AI — reflecting the OSHA General Duty Clause, willful citation, and workers’ compensation dimensions of suppressed PPE and fall protection deficiency detection.

2. BIM clash detection visualisation injection (Autodesk BIM 360 AI, Trimble AI)

BIM clash detection visualisation AI processes screenshots of Autodesk Navisworks clash detection report interfaces, Autodesk BIM 360/Autodesk Construction Cloud clash detection dashboards, Trimble Connect model coordination visualisation displays, and Revit multi-discipline model coordination view screenshots submitted through AI-assisted clash priority ranking, coordination responsibility assignment, and design resolution recommendation tools that MEP engineers, GC BIM coordinators, and specialty subcontractor foremen use to identify structural-MEP clashes, HVAC-sprinkler clearance conflicts, electrical conduit-structural steel interferences, and plumbing-architectural finish space violations in building information models before field installation, generating clash priority scores, coordination responsibility assignments, and resolution recommendation records that govern which coordination issues receive expedited pre-installation resolution versus which are deferred for field coordination, with AIA A201 general conditions coordination responsibility, professional engineer stamped drawing compliance, NFPA 13 sprinkler clearance, and structural PE stamp malpractice liability consequences. Autodesk BIM 360 and Autodesk Construction Cloud AI are deployed on more than one million construction projects, with AI-assisted clash detection and coordination tools processing multi-discipline BIM model visualisation screenshots at commercial office, healthcare, data centre, and infrastructure project sites where MEP coordination complexity generates hundreds to thousands of active clash items per project that BIM coordinators must prioritise, assign, and resolve before field installation crews begin mechanical, electrical, and plumbing rough-in work. Trimble Connect AI processes BIM model coordination visualisation screenshots at international general contractors deploying Trimble’s construction management ecosystem, with AI-assisted clash detection tools integrated into project coordination workflows that govern MEP trade sequencing, prefabrication planning, and subcontractor installation responsibility with AIA A201 contractor coordination obligation and design professional liability dimensions.

The adversarial injection surface is the BIM clash detection visualisation screenshot submission pathway: Autodesk Navisworks or BIM 360 clash detection dashboard screenshots and Trimble Connect model coordination visualisation images submitted through AI-assisted clash priority ranking and coordination responsibility assignment tools for AI MEP clash severity classification, NFPA 13 clearance adequacy determination, and structural interference priority assessment. An adversarially crafted Autodesk BIM 360 clash detection screenshot — in which pixel perturbations applied to the HVAC ductwork-sprinkler branch line interference indicator, the electrical conduit-structural beam clearance measurement display, or the plumbing pipe-concrete pour conflict visual marker on a Navisworks clash detection report interface screenshot cause the AI to classify a critical structural-MEP interference as a low-priority acceptable tolerance clash when the actual screenshot documents an NFPA 13 sprinkler branch line clearance violation or a structural PE-stamped connection zone interference requiring design engineer resolution before installation — can suppress a high-priority clash identification that would otherwise generate an RFI for design professional resolution, preventing field installation of incompatible MEP configurations that require destructive correction after concrete is poured or walls are closed. In BIM coordination workflows where AI-assisted clash prioritisation tools process hundreds of clash items per project per week to identify the critical-path subset requiring expedited resolution before the following week’s field installation activity, adversarial suppression of a structural-HVAC or HVAC-sprinkler clash severity score allows a field-impactful coordination conflict to proceed to installation without the RFI and design resolution workflow that would have prevented it.

The regulatory and liability consequences of adversarially suppressed BIM clash detection in construction coordination AI span AIA A201 contractor coordination, NFPA 13 life safety, and structural PE malpractice dimensions. AIA Document A201-2017 (General Conditions of the Contract for Construction) Article 3.7.4 requires the contractor to take field measurements and verify conditions and report errors, inconsistencies, or omissions discovered; adversarial manipulation of BIM clash detection AI tools that suppresses coordination issue identification and allows incompatible MEP installations creates an Article 3.7.4 contractor coordination breach that exposes the GC to owner back-charge claims for destructive correction costs and schedule delay liquidated damages. NFPA 13 (Standard for the Installation of Sprinkler Systems) specifies clearance requirements between sprinkler branch lines and obstructions including HVAC ductwork, structural elements, and ceiling systems; adversarial suppression of a BIM clash detection AI identification of an NFPA 13 clearance violation that proceeds to field installation creates an as-installed NFPA 13 non-compliance that must be corrected before the authority having jurisdiction will issue a certificate of occupancy, with correction costs borne by the responsible trade contractor under AIA A201 warranty obligations. Structural PE stamp liability arises when AI-suppressed clash detection allows MEP penetrations, support anchors, or prefabricated module installations that violate the structural engineer of record’s stamped and sealed connection zone specifications, creating professional liability exposure for the structural EOR if the as-installed condition differs from the PE-stamped design without RFI-documented design professional approval. Threshold: 60 for BIM clash detection visualisation AI — reflecting the AIA A201 coordination breach, NFPA 13 life safety, and structural PE malpractice dimensions of suppressed MEP interference detection.

3. Construction progress photograph injection (OpenSpace AI, Buildots AI)

Construction progress photograph AI processes 360-degree panoramic progress scan images captured by OpenSpace AI helmet-mounted and tripod-mounted 360-degree cameras documenting more than 600 million square feet of construction space, Buildots AI helmet camera images from Balfour Beatty, Skanska, and Bouygues Construction project sites, Doxel AI construction site scan images, and general contractor superintendent and project manager progress documentation photographs submitted through AI-assisted schedule adherence monitoring, activity completion percentage estimation, and deviation alert generation tools that GC project managers, construction managers, and owners use for AIA G702/G703 payment application progress certification, earned value management calculations, schedule delay analysis, and subcontractor performance evaluation with milestone payment and liquidated damages contractual consequences. OpenSpace AI’s computer vision platform processes 360-degree construction site scan images to generate AI-assisted progress tracking reports that document construction activity completion status, structural and MEP installation sequence adherence, and schedule deviation indicators; its outputs are used by general contractors to support AIA payment application progress certifications and by construction managers to evaluate GC schedule performance in construction management at-risk project delivery structures. Buildots AI deploys construction progress monitoring at ENR Top 10 international general contractor project sites, generating AI-assisted schedule adherence scores and deviation alerts from helmet camera progress images that GC operations directors use for early warning of schedule slippage requiring recovery action, project manager performance evaluation, and liquidated damages risk assessment in fixed-price contract delivery structures.

The adversarial injection surface is the 360-degree construction progress photograph submission pathway: OpenSpace AI 360-degree site scan images and Buildots AI helmet camera progress photographs submitted through AI-assisted schedule deviation detection and activity completion estimation tools for AI construction activity classification, installation sequence adherence scoring, and progress milestone certification support. An adversarially crafted OpenSpace AI 360-degree progress scan image — in which pixel perturbations applied to the structural steel installation completion indicator, the MEP rough-in progress status marker, or the drywall enclosure completion visual cue in a 360-degree construction site scan cause the AI to classify a construction zone as substantially complete with activity sequences meeting schedule milestone criteria when the actual image documents incomplete structural steel connections, uninstalled MEP rough-in items, or open wall cavities failing the schedule milestone completion standard required for the AIA G702 pay application certification — can suppress a schedule deviation alert that would otherwise trigger project manager recovery action or architect-of-record pay application certification review, allowing an overstated progress certification to support a GC pay application that the owner would not have certified if the AI-suppressed schedule deviation had surfaced in the progress report. In construction management at-risk project delivery structures where AI-assisted progress monitoring tools generate the progress documentation that construction managers use to evaluate GC schedule performance and certify AIA pay applications, adversarial suppression of schedule deviation scores concentrates the financial consequences of overstated progress certifications in the owner’s contract position.

The regulatory and contractual consequences of adversarially suppressed schedule deviation detection in construction progress monitoring AI span AIA payment application certification fraud, liquidated damages contract performance, and construction lien law dimensions. AIA G702 Application and Certificate for Payment requires the architect to certify that the work has progressed to the point indicated and the quality of the work is in accordance with the contract documents; adversarial manipulation of AI progress monitoring tools that OpenSpace or Buildots generate as inputs to architect payment certification decisions creates an AI-assisted certification fraud exposure if the AI-suppressed schedule deviation indicator would have caused the architect to withhold or reduce the certification amount. AIA A201-2017 Article 9.8.1 defines substantial completion as the stage in the progress of the work when the work or designated portion thereof is sufficiently complete in accordance with the contract documents so that the owner can occupy or utilise the work for its intended use; AI-suppressed progress monitoring that overstates work completion status and accelerates the owner’s substantial completion determination may affect retainage release timing, warranty period commencement, and liquidated damages cessation with direct financial consequences. Construction contracts at commercial building, data centre, and infrastructure project sites routinely include liquidated damages provisions for schedule overrun; adversarially crafted progress photographs that suppress schedule deviation alerts delay GC corrective action and may increase the ultimate schedule overrun, compounding the liquidated damages liability that the AI-suppressed early warning would have mitigated. Threshold: 65 for construction progress photograph AI — reflecting the AIA pay application certification, liquidated damages, and owner financial loss dimensions of suppressed schedule deviation detection.

4. Building envelope inspection photograph injection (Procore AI, Matterport AI)

Building envelope inspection photograph AI processes exterior wall inspection photographs, window and curtainwall installation documentation images, roofing membrane and flashing installation inspection photographs, and waterproofing application documentation images submitted through Procore AI quality management tools, Matterport AI 3D digital twin building condition assessment platforms, and specialty envelope inspection AI tools that extract water intrusion indicator classifications, flashings deficiency flags, sealant application gap identifications, thermal barrier continuity disruption scores, and IBC Chapter 14 exterior wall compliance assessments from building envelope inspection photograph inputs, generating quality deficiency records, contractor back-charge documentation, warranty claim support records, and certificate of occupancy (CO) inspection preparation reports that GC quality managers, building envelope consultants, and owners use for subcontractor work acceptance, building department CO inspection preparation, and building warranty documentation with IBC, local building department, and contractor warranty liability consequences. Matterport AI processes 3D digital twin capture images from construction and renovation projects through AI-assisted condition assessment tools that building owners, property managers, construction insurers, and litigation support professionals use for building condition documentation, construction defect identification, and pre-litigation evidence preservation; its AI-generated condition assessments are used in construction defect arbitration and litigation proceedings as documentary evidence of as-built conditions. Procore AI integrates building quality inspection photograph analysis into its construction management platform through AI-assisted punch list generation, quality observation recording, and contractor deficiency documentation tools that QC managers at commercial GCs use to manage the final inspection and CO preparation process.

The adversarial injection surface is the building envelope inspection photograph submission pathway: exterior wall, window, roofing, and waterproofing inspection photographs submitted through Procore AI quality management tools or Matterport AI building condition assessment platforms for AI water intrusion indicator classification, flashings deficiency scoring, and IBC Chapter 14 exterior wall compliance assessment. An adversarially crafted building envelope inspection photograph — in which pixel perturbations applied to the window-to-wall interface sealant gap visual marker, the roof membrane termination flashing deficiency indicator, or the exterior wall water infiltration staining pattern in a building envelope inspection photograph cause the AI to classify an envelope assembly with IBC Chapter 14-non-compliant or building-science-deficient installation as meeting exterior wall performance standard criteria when the actual image documents a water intrusion pathway or IBC-non-compliant exterior wall assembly installation — can suppress an envelope deficiency identification that would otherwise generate a contractor back-charge record and warranty remediation work order, allowing a water-infiltration-susceptible building envelope assembly to receive QC acceptance documentation that the GC would not have provided if the AI-suppressed deficiency indicator had surfaced in the quality inspection record. In high-rise commercial, healthcare, and data centre project completions where AI-assisted building envelope inspection tools process thousands of cladding, window, and roofing inspection photographs in the compressed final inspection and CO preparation timeline, adversarial suppression of envelope deficiency scores allows latent water intrusion pathways to proceed to CO and building occupancy without the remediation documentation that would have been required under the GC’s quality management plan.

The regulatory and liability consequences of adversarially suppressed building envelope deficiency detection span IBC Chapter 14 exterior wall compliance, local building department certificate of occupancy, and contractor warranty and latent defect liability dimensions. IBC Chapter 14 (Exterior Walls) establishes performance requirements for exterior wall assemblies including weather protection, water-resistive barrier installation, flashing requirements at penetrations and transitions, and drainage provisions; adversarial manipulation of building envelope inspection AI tools that suppresses a Chapter 14 non-compliance identification before building department CO inspection creates a building code violation that the authority having jurisdiction may identify at CO inspection, requiring correction before certificate of occupancy issuance and potentially triggering building department stop-work authority for the affected envelope assembly. Contractor warranty obligations under AIA A201-2017 Article 3.5 (Warranty) require the contractor to warrant that materials and equipment furnished under the contract will be of good quality and new, that the work will be free from defects, and that the work will conform to the requirements of the contract documents; adversarial suppression of building envelope AI deficiency detection that allows a non-conforming envelope installation to receive QC acceptance documentation creates a warranty claim exposure that persists for the full AIA warranty period (one year from the date of substantial completion) and for the latent defect period under applicable state construction defect statutes, which extend to ten years in California (CCP §337.15) and similar periods in other states. Matterport AI 3D digital twin building condition assessment records are increasingly used in construction defect litigation as documentary evidence of building envelope conditions at the time of inspection; adversarial manipulation of Matterport AI condition assessment tools that suppresses water intrusion indicator identification creates a litigation evidence integrity dimension to the adversarial injection exposure where the manipulated AI assessment records may be presented as exculpatory evidence in construction defect proceedings. Threshold: 60 for building envelope inspection AI — reflecting the IBC Chapter 14 compliance, certificate of occupancy, and contractor warranty and latent defect liability dimensions of suppressed envelope deficiency detection.

Integration: construction and BIM AI image ingestion with Glyphward pre-scan

Construction and BIM AI image ingestion flows from Procore and Smartvid.io jobsite safety camera photograph APIs, Autodesk BIM 360 and Trimble Connect clash detection screenshot channels, OpenSpace and Buildots progress monitoring 360-degree photograph interfaces, and Procore and Matterport building envelope inspection photograph portals into safety observation AI, clash priority ranking AI, schedule deviation detection AI, and building condition assessment AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to OSHA safety program records, BIM coordination responsibility assignments, AIA pay application certifications, or building envelope QC acceptance records:

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"

# Construction & BIM AI — OSHA 29 CFR Part 1926; AIA A201; NFPA 13;
# IBC Chapter 14. Suppression of PPE violations, MEP clashes, schedule
# deviations, and envelope deficiencies create OSHA willful citation,
# contractor coordination breach, and warranty liability consequences.
THRESHOLD_SAFETY_CAMERA  = 55  # Procore/Smartvid.io; OSHA CIP; GD Clause
THRESHOLD_CONSTRUCTION_AI = 60  # BIM clash, progress monitoring, envelope inspection

# Progress photograph threshold is elevated — AIA pay app fraud exposure
THRESHOLD_PROGRESS_PHOTO = 65  # OpenSpace/Buildots; AIA G702; liquidated damages


class ConstructionAIContext(str, Enum):
    SAFETY_CAMERA      = "safety_camera"       # Procore, Smartvid.io, Doxel
    BIM_CLASH          = "bim_clash"           # Autodesk BIM 360, Trimble
    PROGRESS_PHOTO     = "progress_photo"      # OpenSpace, Buildots
    ENVELOPE_INSPECT   = "envelope_inspect"    # Procore, Matterport


def threshold_for(context: ConstructionAIContext) -> int:
    if context == ConstructionAIContext.SAFETY_CAMERA:
        return THRESHOLD_SAFETY_CAMERA
    if context == ConstructionAIContext.PROGRESS_PHOTO:
        return THRESHOLD_PROGRESS_PHOTO
    return THRESHOLD_CONSTRUCTION_AI


async def scan_construction_ai_image(
    image_path: str | Path,
    context: ConstructionAIContext,
    project_id_hash: str,   # SHA-256 of Procore/Autodesk project ID
    contractor_ref: str,    # e.g. "GC-TRN-2026-0441", "SUB-MECH-887234"
    observation_id: str,    # safety obs ID, clash ID, scan session ID, inspection ID
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a construction or BIM AI image for adversarial injection payloads
    before forwarding to safety observation scoring, clash priority ranking,
    progress deviation detection, or building envelope condition assessment.

    Raises AdversarialConstructionAIImageError if score meets threshold:
      - SAFETY_CAMERA:    threshold 55; OSHA 29 CFR Part 1926 willful citation;
                          General Duty Clause; workers' compensation EMR
      - BIM_CLASH:        threshold 60; AIA A201 Article 3.7.4; NFPA 13;
                          structural PE malpractice; contractor coordination
      - PROGRESS_PHOTO:   threshold 65; AIA G702 pay app certification;
                          liquidated damages; substantial completion milestone
      - ENVELOPE_INSPECT: threshold 60; IBC Chapter 14; CO inspection;
                          AIA A201 Article 3.5 warranty; latent defect statutes
    """
    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": {
                "construction_context": context.value,
                "project_id_hash":      project_id_hash,
                "contractor_ref":       contractor_ref,
                "observation_id":       observation_id,
                "client_scan_id":       client_scan_id,
                "image_sha256":         image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "project_id_hash":      project_id_hash,
        "contractor_ref":       contractor_ref,
        "observation_id":       observation_id,
        "construction_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_construction_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialConstructionAIImageError(
            f"Construction AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"project={project_id_hash} contractor={contractor_ref}"
        )
    return result


async def write_construction_audit_record(record: dict) -> None:
    """Persist audit record to construction project compliance store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialConstructionAIImageError(Exception):
    """Raised when a construction or BIM AI image exceeds the adversarial injection threshold."""
    pass

Call scan_construction_ai_image() with ConstructionAIContext.SAFETY_CAMERA before forwarding Procore or Smartvid.io jobsite safety photographs to AI PPE compliance and fall protection deficiency classification tools — the integration point where adversarial suppression of an OSHA-reportable condition creates General Duty Clause willful citation exposure, with observation_id set to the Procore safety observation identifier linking the Glyphward scan record to the specific OSHA 300 log recordkeeping event. Call with ConstructionAIContext.BIM_CLASH for Autodesk BIM 360 or Trimble Connect clash detection dashboard screenshots before AI MEP clash severity classification and NFPA 13 clearance assessment, preserving image_sha256 as the forensic anchor for AIA A201 contractor coordination audit and NFPA 13 inspection compliance documentation. Call with ConstructionAIContext.PROGRESS_PHOTO for OpenSpace AI or Buildots AI 360-degree progress scan images before AI schedule deviation detection and activity completion estimation, with observation_id set to the progress scan session identifier for AIA G702 pay application certification audit trail linkage. Call with ConstructionAIContext.ENVELOPE_INSPECT for Procore quality management or Matterport AI building condition assessment inspection photographs before AI water intrusion indicator classification and IBC Chapter 14 compliance scoring, with contractor_ref encoding the envelope subcontractor scope for AIA A201 warranty and latent defect liability documentation purposes. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Indirect prompt injection via image — foundational attack pattern underlying all four construction and BIM AI injection surfaces; covers how adversarial pixel-level perturbations cause AI misclassification without detectable visual artefacts at human review resolution, directly applicable to jobsite safety camera photograph and building envelope inspection photograph injection.
• Vision-language model security — technical architecture of adversarial image attacks against vision-language models including pixel perturbation classes applicable to BIM clash detection visualisation screenshot injection and construction progress monitoring photograph manipulation.
• Manufacturing quality inspection AI prompt injection — industrial AI injection context covering the broader class of physical inspection photograph manipulation applicable to building envelope inspection AI injection and construction progress quality documentation adversarial attacks.
• Free tier — 10 scans/day, no card required — start scanning construction and BIM AI images at development volumes before committing to a production plan.