Prompt injection in public health disease surveillance and epidemiology AI

Four adversarial injection surfaces in public health disease surveillance and epidemiology AI

1. Disease outbreak surveillance image injection (21 USC §264, PHSA §319)

Disease outbreak surveillance AI processes epidemiological map visualization images displaying geographic case distribution patterns and outbreak cluster hotspot heatmaps, outbreak case count trend display images showing daily and cumulative case trajectories, geographic spread progression animation display images showing outbreak boundary expansion, epidemiological curve (epi-curve) images displaying case onset date distributions and outbreak trajectory classification indicators, outbreak doubling time and reproductive number (R₀) display images, and emergency operations center activation threshold indicator display images — processed by CDC BioSense Platform AI at 1,600 or more hospital facilities reporting syndromic surveillance data covering 200 million or more annual emergency department visits through AI-assisted syndromic signal escalation and outbreak threshold determination tools; HealthMap AI at 100 or more countries monitoring 250 or more data sources in 30 or more languages through AI-assisted real-time disease event detection and geographic outbreak cluster mapping tools; BlueDot AI at 137 countries disease surveillance through AI-assisted disease event signal detection and geographic spread modeling tools; and Palantir Foundry public health AI at 30 or more state and national public health agency deployments including NHS England through AI-assisted epidemiological data integration and outbreak intelligence platform tools that process outbreak visualization images from state health department reporting systems, hospital syndromic surveillance feeds, and CDC surveillance data aggregation platforms — extracting outbreak escalation threshold determinations and CDC Emergency Operations Center activation recommendations from outbreak map and epi-curve visualization image inputs in AI-assisted real-time disease outbreak detection pipelines.

The adversarial injection surface is the outbreak epidemiological map visualization image or epi-curve display image submission pathway: CDC BioSense Platform AI or HealthMap AI outbreak visualization images submitted through AI-assisted outbreak signal escalation and emergency response threshold determination tools for AI outbreak classification record generation and CDC Emergency Operations Center activation input. An adversarially crafted outbreak map image or epi-curve display image — in which pixel perturbations applied to the outbreak case count escalation indicator display region of an epidemiological map image, the epi-curve trajectory steepening and acceleration visual marker of a case onset date distribution display, or the geographic cluster density threshold exceedance indicator in an outbreak heatmap visualization cause the AI to classify a developing disease outbreak as a contained local event not meeting CDC Emergency Operations Center activation criteria when the actual case count trajectory and geographic spread pattern evidences outbreak escalation indicators that would otherwise generate an outbreak emergency response referral — can suppress an outbreak escalation signal that would otherwise generate a CDC Emergency Operations Center activation, a public health emergency response mobilization, and a PHSA §319 Public Health Emergency declaration process initiation. In CDC outbreak monitoring operations where BioSense Platform AI processes outbreak map and epi-curve visualization images from 1,600 or more hospital syndromic surveillance feeds covering 200 million or more annual emergency department visits without individual human epidemiologist visual examination of every AI-processed surveillance visualization at the pixel level before the AI classification governs the outbreak escalation determination, adversarial suppression of outbreak escalation indicators creates PHSA §319, 21 USC §264, and WHO IHR 2005 Article 6 notification obligation dimensions of potentially catastrophic public health severity.

The 21 USC §264, PHSA §319, HHS 42 CFR Part 70, and WHO IHR 2005 consequences of adversarially suppressed outbreak escalation classification in disease surveillance AI span 21 USC §264 Public Health Service Act Section 361 quarantine and inspection authority granted to CDC to make and enforce regulations necessary to prevent the introduction, transmission, or spread of communicable diseases from foreign countries into the United States or from one state to another through measures including apprehension, detention, and conditional release of individuals; PHSA §319 (42 USC §247d) Public Health Emergency declaration authority enabling the HHS Secretary to declare a public health emergency and to deploy Strategic National Stockpile (SNS) medical countermeasures, issue emergency use authorizations (EUA) under PHSA §564, waive certain Medicare and Medicaid requirements under PHSA §1135, and coordinate federal emergency response resources including the National Disaster Medical System (NDMS); HHS 42 CFR Part 70 interstate quarantine regulations establishing notification and control requirements applicable to the spread of specified communicable diseases between states including COVID-19, viral hemorrhagic fevers, plague, smallpox, cholera, and other quarantinable communicable diseases listed in Executive Order 13295 as amended; and WHO International Health Regulations (2005) Article 6 notification obligations requiring States Parties to notify WHO of any event which may constitute a public health emergency of international concern within 24 hours of assessment using the Annex 2 decision instrument, with WHO IHR 2005 Article 44 obligations for States Parties to collaborate and mutually assist one another in technical surveillance and response activities. PHSA §319 Public Health Emergency declarations have historically enabled federal mobilization of billions of dollars in emergency supplemental appropriations, Strategic National Stockpile deployment, and coordinated federal-state public health emergency response resources; adversarial delay of a PHSA §319 declaration by suppressing outbreak escalation signals in CDC AI surveillance tools would delay the federal public health emergency response at precisely the moment of highest consequence for reducing outbreak morbidity and mortality. Threshold: 55 for outbreak surveillance AI — reflecting 21 USC §264 quarantine authority, PHSA §319 Public Health Emergency declaration, 42 CFR Part 70 interstate notification, and WHO IHR 2005 Article 6 PHEIC notification obligation dimensions.

2. Genomic epidemiology sequencing injection (42 USC §243, MMWR reporting, WHO IHR 2005)

Genomic epidemiology AI processes whole-genome sequencing (WGS) phylogenetic tree visualization images displaying clade assignment and evolutionary lineage patterns for outbreak pathogen characterization, pathogen variant classification display images showing variant designation and phenotypic property assignment, genomic transmission chain network graph images displaying inferred transmission linkage patterns between sequenced cases, pathogen sequence alignment display images from Illumina sequencing platform outputs showing nucleotide position variation patterns, GISAID EpiCoV database sequence submission and clade designation display images, and nextstrain.org phylogenetic tree and geographic spread visualization images — processed by Johns Hopkins CSSE AI across 200 or more independent data sources daily through AI-assisted disease data aggregation and genomic epidemiology visualization tools; Palantir Foundry public health AI at 30 or more state and national public health agencies including NHS England and multiple US state health departments through AI-assisted genomic epidemiology and outbreak intelligence platform tools processing WGS phylogenetic visualization images from Illumina sequencing pipelines and GISAID database feeds; and HealthMap AI and BlueDot AI through AI-assisted genomic epidemiology surveillance and outbreak characterization tools that process pathogen sequence visualization images from international genomic surveillance networks — extracting outbreak strain characterization classifications and transmission pathway identification determinations from WGS phylogenetic tree and transmission chain network graph visualization image inputs in AI-assisted genomic epidemiology outbreak characterization pipelines.

The adversarial injection surface is the phylogenetic tree visualization image or transmission chain network graph image submission pathway: Johns Hopkins CSSE AI or Palantir Foundry public health AI genomic epidemiology visualization images submitted through AI-assisted outbreak strain characterization and transmission pathway identification tools for AI genomic epidemiology classification record generation and public health variant response protocol input. An adversarially crafted phylogenetic tree visualization image or transmission chain network graph image — in which pixel perturbations applied to the clade boundary and clade assignment indicator display region of a WGS phylogenetic tree image, the transmission chain directionality arrow and linkage confidence visual marker of a transmission network graph, or the novel variant emergence indicator display in a pathogen sequence alignment visualization cause the AI to misassign outbreak strains to known benign clades, alter the inferred directionality of transmission chains between sequenced cases, or conceal the emergence of a novel variant with potentially altered transmissibility or immune escape properties from AI classification systems when the actual phylogenetic tree topology and sequence data evidences novel clade emergence, accelerated transmission spread, or variant-specific phenotypic signals that would otherwise generate a WHO-reportable variant of concern classification — can suppress a novel variant emergence indicator that would otherwise generate a WHO IHR 2005 Article 6 PHEIC notification, a CDC variant classification and surveillance expansion response, and an MMWR outbreak investigation report. In public health genomic surveillance operations where Johns Hopkins CSSE AI or Palantir Foundry AI processes WGS phylogenetic tree visualization images from global genomic surveillance network data feeds without individual human phylogeneticist pixel-level examination of every AI-processed genomic visualization before the AI classification governs the variant designation and public health response protocol, adversarial suppression of novel variant emergence and transmission chain escalation indicators creates 42 USC §243, MMWR, and WHO IHR 2005 Article 6 dimensions.

The 42 USC §243, MMWR reporting, WHO IHR 2005, and CDC variant classification consequences of adversarially suppressed genomic epidemiology classification span 42 USC §243 CDC cooperative agreement authority enabling cooperative agreements with state and local health departments for disease surveillance, investigation, and control including genomic surveillance network activities; CDC MMWR Morbidity and Mortality Weekly Report reporting requirements applicable to notifiable disease case reporting and outbreak investigation findings including genomic epidemiology characterization results for novel pathogen variants with public health significance; WHO International Health Regulations (2005) Article 6 State Party notification obligations requiring notification to WHO of events including novel influenza virus subtypes, SARS, viral hemorrhagic fevers, and other events that may constitute a PHEIC, with the Article 6 assessment framework using the WHO Annex 2 decision instrument requiring notification of events with significant public health consequences; and WHO Integrated Disease Surveillance and Response (IDSR) technical guidelines applicable to national public health surveillance system event detection and reporting standards including genomic epidemiology characterization of outbreak pathogens. The COVID-19 pandemic demonstrated the global public health consequences of delayed genomic surveillance and variant classification: SARS-CoV-2 Omicron was identified through genomic surveillance in southern Africa and classified as a Variant of Concern by WHO on 26 November 2021, triggering immediate international travel restriction responses; adversarial manipulation of genomic epidemiology AI that suppresses novel variant emergence indicators or alters clade assignment classifications would delay the WHO variant classification process and the coordinated international public health response it triggers. BlueDot AI’s identification of COVID-19 nine days before the WHO announcement demonstrates the early warning value of AI-assisted genomic and epidemiological data analysis; adversarially compromised genomic AI that suppresses emergence signals would eliminate precisely the early warning advantage that AI-assisted genomic surveillance provides over traditional human-review epidemiology. Threshold: 55 for genomic epidemiology AI — reflecting 42 USC §243 cooperative surveillance obligations, MMWR outbreak reporting requirements, WHO IHR 2005 Article 6 PHEIC notification obligations, and CDC variant classification process integrity dimensions.

3. Syndromic surveillance injection (HIPAA §164.512(b) public health reporting)

Syndromic surveillance AI processes emergency department chief complaint category distribution display images showing ILI (influenza-like illness), respiratory illness, gastrointestinal illness, and injury category clustering patterns from hospital EHR BioSense feed data, school and workplace absenteeism rate trend display images from syndromic absenteeism monitoring systems, over-the-counter pharmaceutical sales trend display images from pharmacy syndromic surveillance data including antidiarrheal, antinausea, fever reducer, and cough/cold product category sales anomaly detection, 911 call category distribution display images from emergency dispatch syndromic surveillance data, and poison control center call category trend display images — processed by CDC BioSense Platform AI at 1,600 or more hospital facilities covering 200 million or more annual emergency department visits through AI-assisted ED visit pattern analysis and syndromic outbreak signal detection tools; state and local health department public health AI platforms receiving BioSense data feeds through Palantir Foundry public health AI at 30 or more agency deployments and through CDC-funded cooperative agreement public health informatics systems; and HealthMap AI and BlueDot AI through AI-assisted syndromic surveillance signal monitoring and early outbreak detection tools that process syndromic data visualization images from hospital, pharmacy, and absenteeism surveillance data feeds — extracting syndromic outbreak precursor signal classifications and public health emergency threshold exceedance determinations from ED chief complaint, absenteeism, and pharmaceutical sales trend display image inputs in AI-assisted early outbreak detection pipelines before laboratory confirmation of outbreak etiology.

The adversarial injection surface is the ED chief complaint category distribution display image or pharmaceutical sales trend display image submission pathway: CDC BioSense Platform AI or state health department AI syndromic surveillance trend display images submitted through AI-assisted syndromic signal escalation and outbreak precursor detection tools for AI syndromic threshold exceedance classification record generation and public health alert or response initiation input. An adversarially crafted ED chief complaint category distribution display image or OTC pharmaceutical sales trend display image — in which pixel perturbations applied to the ILI category rate escalation indicator display region of an ED visit distribution image, the geographical clustering concentration visual marker of an absenteeism rate heatmap, or the threshold exceedance level indicator display in a pharmaceutical sales trend visualization cause the AI to classify an emerging syndromic outbreak signal as a normal seasonal variation not meeting syndromic alert threshold criteria when the actual ED visit distribution, absenteeism rate, and pharmaceutical sales trend data evidences unusual clustering signals, threshold exceedance indicators, or outbreak precursor patterns that would otherwise generate a public health alert, an epidemiological field investigation initiation, and a PHSA §319 Public Health Emergency pre-notification escalation. In state and local public health operations where BioSense Platform AI processes ED chief complaint and syndromic surveillance trend images from 1,600 or more hospital facilities without individual human epidemiologist visual examination of every AI-processed syndromic visualization at the pixel level before the AI classification governs the syndromic alert threshold determination, adversarial suppression of syndromic outbreak precursor signals creates HIPAA §164.512(b), CSTE reportable condition, and CDC BioSense platform data integrity dimensions.

The HIPAA §164.512(b), CSTE reportable condition, 42 CFR Part 70, and CDC BioSense platform consequences of adversarially suppressed syndromic surveillance signal classification span HIPAA §164.512(b) public health reporting exemption authorizing covered entities to disclose protected health information to authorized public health authorities including CDC and state and local health departments for purposes of preventing or controlling disease, injury, or disability without individual authorization, establishing the legal basis for hospital EHR syndromic surveillance data flows to CDC BioSense and state health department syndromic systems; Council of State and Territorial Epidemiologists (CSTE) reportable condition notification requirements establishing the NNDSS (National Notifiable Diseases Surveillance System) notifiable disease reporting standards applicable to healthcare providers and laboratories reporting confirmed and probable cases of over 120 nationally notifiable conditions to state health departments and from state departments to CDC; 42 CFR Part 70 interstate quarantine notification requirements applicable to events involving the interstate spread of communicable diseases including the notification timelines for quarantinable disease events at US ports of entry and between states; and CDC BioSense Platform cooperative agreement data use requirements applicable to state and local health department BioSense data sharing, analysis, and use in public health surveillance and outbreak investigation activities. The CSTE 2021 revision of nationally notifiable conditions added COVID-19 as a nationally notifiable condition; adversarial delay of syndromic outbreak signal detection through corrupted syndromic surveillance AI visualization analysis creates the conditions for delayed CSTE reportable case notification and delayed CDC NNDSS case ascertainment for emerging outbreak conditions. PHSA §319 Public Health Emergency pre-notification escalation is informed by syndromic surveillance data trends from the CDC BioSense system; adversarially suppressed syndromic threshold exceedance signals would delay the information flow from BioSense syndromic AI analysis to CDC Emergency Operations Center pre-notification briefings that inform the HHS Secretary’s PHSA §319 declaration assessment. Threshold: 60 for syndromic surveillance AI — reflecting HIPAA §164.512(b) public health reporting accuracy obligations, CSTE reportable condition notification timeliness requirements, 42 CFR Part 70 interstate quarantine notification, and the CDC BioSense platform data integrity dimensions that make a more conservative threshold appropriate than for outbreak surveillance AI.

4. Pathogen biocontainment image injection (42 CFR Part 73 select agents)

Pathogen biocontainment AI processes laboratory biosafety level (BSL) classification display images showing BSL-1 through BSL-4 containment zone status and access control indicator patterns, pathogen containment status indicator images showing positive pressure suit area integrity, HEPA filtration system status, and primary and secondary containment barrier indicator patterns, select agent inventory reconciliation display images showing quantity, location, and transfer authorization record status for 42 CFR Part 73 regulated biological select agents and toxins, BSL-3 and BSL-4 facility environmental monitoring camera feed images showing laboratory access control, pressure differential indicator, and biosafety cabinet status patterns, and Federal Select Agent Program (FSAP) inspection compliance status display images — processed by Palantir Foundry public health AI at public health agency and biodefense laboratory deployments processing biosafety compliance and biocontainment monitoring data through AI-assisted laboratory security and containment status analysis tools; Metabiota AI at epidemic risk modeling operations for 100 or more insurance and reinsurance clients through AI-assisted biosafety risk assessment and epidemic probability modeling tools that process pathogen containment status and laboratory biosafety classification display images; and CDC Federal Select Agent Program inspection and compliance monitoring systems processing select agent inventory reconciliation display images through AI-assisted FSAP compliance tracking and inventory integrity monitoring tools — extracting biocontainment compliance status determinations and BSL-3/BSL-4 containment integrity classifications from pathogen containment monitoring display image inputs in AI-assisted laboratory biosafety compliance and pathogen security monitoring pipelines.

The adversarial injection surface is the BSL classification display image or pathogen containment status indicator image submission pathway: Palantir Foundry public health AI or Metabiota AI biocontainment monitoring display images submitted through AI-assisted biosafety compliance and containment integrity monitoring tools for AI biocontainment status classification record generation and FSAP compliance reporting input. An adversarially crafted BSL-4 containment status indicator image or select agent inventory reconciliation display image — in which pixel perturbations applied to the containment barrier integrity indicator display region of a BSL-4 facility monitoring image, the positive pressure differential status visual marker of a BSL-3 laboratory pressure monitoring display, or the select agent inventory quantity reconciliation discrepancy indicator in an FSAP inventory status display cause the AI to classify an active containment anomaly or select agent inventory discrepancy as a compliant containment event not meeting FSAP incident reporting criteria when the actual containment monitoring data and inventory reconciliation records evidence a containment integrity anomaly, a pressure differential deviation, or a select agent inventory discrepancy that would otherwise generate an immediate FSAP notification, a biosafety officer incident investigation initiation, and a CDC/USDA Federal Select Agent Program regulatory response. In biosafety laboratory monitoring operations where AI processes BSL-3 and BSL-4 containment status and select agent inventory display images without continuous individual human biosafety officer visual examination of every AI-processed monitoring display at the pixel level before the AI classification governs the containment compliance status determination, adversarial suppression of containment breach and select agent inventory discrepancy indicators creates 42 CFR Part 73, HHS/USDA FSAP, and biosecurity consequence dimensions of the highest severity in the entire public health surveillance domain.

The 42 CFR Part 73, HHS Federal Select Agent Program, USDA APHIS, and biosecurity consequences of adversarially suppressed pathogen biocontainment classification span 42 CFR Part 73 (HHS regulations for select agents and toxins) establishing registration requirements, inventory accountability standards, theft, loss, and release reporting requirements within 24 hours under 42 CFR §73.19(a), and security risk assessment (SRA) requirements for all individuals with access to select agents; 9 CFR Part 121 (USDA APHIS regulations for overlap select agents) establishing parallel registration, inventory, and incident reporting requirements for overlap agents regulated by both HHS and USDA; HHS Federal Select Agent Program and USDA Agricultural Select Agent Program joint inspection authority under 42 CFR §73.9 requiring registered entities to allow FSAP inspectors access to all areas of the facility and to all relevant records; 18 USC §175 biological weapons prohibition creating criminal penalties of up to life imprisonment for any person who knowingly develops, produces, stockpiles, transfers, acquires, retains, or possesses any biological agent, toxin, or delivery system for use as a weapon; and 18 USC §175b creating criminal penalties of up to 10 years imprisonment for possession of a biological agent or toxin by a restricted person. 42 CFR §73.19(a) imposes a 24-hour reporting obligation for any theft or loss of a select agent or toxin, or any release of a select agent or toxin that may pose a risk to the health and safety of employees or the public or to animal or plant health; adversarial suppression of containment breach or inventory discrepancy indicators in biocontainment AI would delay the 42 CFR §73.19(a) reporting timeline, potentially enabling a select agent release event to go unreported beyond the 24-hour regulatory notification window. The CDC FSAP has identified more than 200 registered entities that possess select agents in the United States; the consequences of an adversarially induced failure in biocontainment AI monitoring at a BSL-4 facility handling Tier 1 select agents such as Ebola virus, Marburg virus, or variola major extend beyond regulatory violation to public health catastrophe. Threshold: 45 for pathogen biocontainment AI — the lowest threshold in the entire Glyphward product, reserved for the highest-consequence AI monitoring surface where the adversarial false negative consequence is a BSL-3/BSL-4 containment breach or select agent inventory discrepancy going undetected and unreported beyond 42 CFR §73.19(a)’s 24-hour mandatory notification window.

Integration: public health disease surveillance AI image ingestion with Glyphward pre-scan

Public health disease surveillance AI image ingestion flows from CDC BioSense Platform AI outbreak map visualization and epi-curve display image ingestion channels, Johns Hopkins CSSE AI and Palantir Foundry AI genomic phylogenetic tree and transmission chain network graph image processing interfaces, CDC BioSense Platform AI and state health department AI ED chief complaint and syndromic surveillance trend display image analysis systems, and Palantir Foundry public health AI and Metabiota AI BSL-3/BSL-4 biocontainment monitoring and select agent inventory reconciliation display image processing platforms into disease outbreak escalation threshold determination AI, genomic outbreak strain characterization and variant classification AI, syndromic outbreak precursor signal detection AI, and pathogen biocontainment compliance and containment integrity monitoring AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to CDC Emergency Operations Center activation records, WHO IHR 2005 Article 6 notification records, CSTE reportable condition notification records, or FSAP 42 CFR §73.19(a) mandatory incident reporting 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"

# Public health disease surveillance AI — 21 USC §264 quarantine authority; PHSA §319 PHE;
# 42 CFR Part 70 interstate quarantine; 42 USC §243 CDC cooperative agreements;
# MMWR reportable disease; WHO IHR 2005 Art. 6 PHEIC notification;
# HIPAA §164.512(b) public health reporting; CSTE reportable conditions;
# 42 CFR Part 73 Select Agent Program (highest-risk — 24-hr incident reporting §73.19(a)).
THRESHOLD_OUTBREAK_SURVEILLANCE_AI     = 55  # BioSense/HealthMap; 21 USC §264; PHSA §319 PHE
THRESHOLD_GENOMIC_EPIDEMIOLOGY_AI      = 55  # JHU CSSE/Palantir; 42 USC §243; MMWR; WHO IHR 2005
THRESHOLD_SYNDROMIC_SURVEILLANCE_AI    = 60  # BioSense/state dept; HIPAA §164.512(b); CSTE reportable
THRESHOLD_PATHOGEN_BIOCONTAINMENT_AI   = 45  # Palantir/Metabiota; 42 CFR Part 73; §73.19(a) 24hr


class PublicHealthSurveillanceAIContext(str, Enum):
    OUTBREAK_SURVEILLANCE_AI   = "outbreak_surveillance_ai"    # BioSense, HealthMap, BlueDot
    GENOMIC_EPIDEMIOLOGY_AI    = "genomic_epidemiology_ai"     # JHU CSSE, Palantir, HealthMap
    SYNDROMIC_SURVEILLANCE_AI  = "syndromic_surveillance_ai"   # BioSense, state depts, Palantir
    PATHOGEN_BIOCONTAINMENT_AI = "pathogen_biocontainment_ai"  # Palantir, Metabiota, CDC FSAP


def threshold_for(context: PublicHealthSurveillanceAIContext) -> int:
    mapping = {
        PublicHealthSurveillanceAIContext.OUTBREAK_SURVEILLANCE_AI:   THRESHOLD_OUTBREAK_SURVEILLANCE_AI,
        PublicHealthSurveillanceAIContext.GENOMIC_EPIDEMIOLOGY_AI:    THRESHOLD_GENOMIC_EPIDEMIOLOGY_AI,
        PublicHealthSurveillanceAIContext.SYNDROMIC_SURVEILLANCE_AI:  THRESHOLD_SYNDROMIC_SURVEILLANCE_AI,
        PublicHealthSurveillanceAIContext.PATHOGEN_BIOCONTAINMENT_AI: THRESHOLD_PATHOGEN_BIOCONTAINMENT_AI,
    }
    return mapping[context]


async def scan_public_health_surveillance_ai_image(
    image_path: str | Path,
    context: PublicHealthSurveillanceAIContext,
    event_ref_hash: str,          # SHA-256 of outbreak event ID, WGS accession, or FSAP lot number
    surveillance_ref: str,        # e.g. "BIOSENSE-2026-EOC-441", "GISAID-EPI-2026-8819", "FSAP-TIER1-441"
    surveillance_session_id: str, # BioSense monitoring session, GISAID submission batch, FSAP audit session
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a public health disease surveillance AI image for adversarial injection payloads
    before forwarding to disease outbreak escalation threshold determination, genomic
    epidemiology strain characterization and variant classification, syndromic surveillance
    outbreak precursor signal detection, or pathogen biocontainment compliance and
    containment integrity monitoring AI.

    Raises AdversarialPublicHealthSurveillanceAIImageError if score meets threshold:
      - OUTBREAK_SURVEILLANCE_AI:   threshold 55; 21 USC §264; PHSA §319 PHE declaration
      - GENOMIC_EPIDEMIOLOGY_AI:    threshold 55; 42 USC §243; MMWR; WHO IHR 2005 Art. 6
      - SYNDROMIC_SURVEILLANCE_AI:  threshold 60; HIPAA §164.512(b); CSTE reportable conditions
      - PATHOGEN_BIOCONTAINMENT_AI: threshold 45 (LOWEST IN PRODUCT); 42 CFR Part 73; §73.19(a)
    """
    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": {
                "public_health_surveillance_context": context.value,
                "event_ref_hash":                    event_ref_hash,
                "surveillance_ref":                  surveillance_ref,
                "surveillance_session_id":           surveillance_session_id,
                "client_scan_id":                    client_scan_id,
                "image_sha256":                      image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "event_ref_hash":                    event_ref_hash,
        "surveillance_ref":                  surveillance_ref,
        "surveillance_session_id":           surveillance_session_id,
        "public_health_surveillance_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_public_health_surveillance_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialPublicHealthSurveillanceAIImageError(
            f"Public health surveillance AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"event={event_ref_hash} ref={surveillance_ref}"
        )
    return result


async def write_public_health_surveillance_audit_record(record: dict) -> None:
    """Persist audit record to public health surveillance regulatory documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialPublicHealthSurveillanceAIImageError(Exception):
    """Raised when a public health surveillance AI image exceeds the adversarial injection threshold."""
    pass

Call scan_public_health_surveillance_ai_image() with PublicHealthSurveillanceAIContext.OUTBREAK_SURVEILLANCE_AI before forwarding CDC BioSense Platform AI or HealthMap AI outbreak epidemiological map visualization images, epi-curve display images, and geographic case count heatmap images to outbreak escalation threshold determination and Emergency Operations Center activation input AI — with surveillance_ref as the CDC outbreak event identifier for 21 USC §264 quarantine authority, PHSA §319 Public Health Emergency declaration integrity, and WHO IHR 2005 Article 6 PHEIC notification obligation compliance audit trail documentation. Call with PublicHealthSurveillanceAIContext.GENOMIC_EPIDEMIOLOGY_AI for Johns Hopkins CSSE AI or Palantir Foundry public health AI WGS phylogenetic tree visualization images, variant classification display images, and transmission chain network graph images before genomic outbreak strain characterization and variant classification AI, with event_ref_hash as the SHA-256 of the GISAID EpiCoV accession number for 42 USC §243 cooperative surveillance, MMWR outbreak reporting, and WHO IHR 2005 Article 6 PHEIC notification obligation audit trail. Call with PublicHealthSurveillanceAIContext.SYNDROMIC_SURVEILLANCE_AI for CDC BioSense Platform AI or state health department AI ED chief complaint distribution display images, OTC pharmaceutical sales trend images, and absenteeism rate display images before syndromic outbreak precursor signal detection AI, with surveillance_session_id for HIPAA §164.512(b) public health reporting accuracy, CSTE reportable condition notification timeliness, and 42 CFR Part 70 interstate quarantine notification compliance audit trail. Call with PublicHealthSurveillanceAIContext.PATHOGEN_BIOCONTAINMENT_AI for Palantir Foundry public health AI or Metabiota AI BSL-3/BSL-4 containment status display images, select agent inventory reconciliation display images, and FSAP compliance monitoring display images before pathogen biocontainment compliance and containment integrity AI — with threshold 45, the most conservative threshold in the entire Glyphward product, reflecting 42 CFR Part 73 select agent program requirements, 42 CFR §73.19(a) 24-hour mandatory incident reporting obligations, and the BSL-4 containment breach public health catastrophe consequence of a false negative detection miss. Get early access

Coverage matrix

Related questions

Further reading

• FigStep adversarial image injection detection — technical documentation of pixel-level FigStep image injection attacks applicable to outbreak epidemiological map visualization and WGS phylogenetic tree display adversarial manipulation in CDC BioSense AI and Johns Hopkins CSSE AI genomic surveillance pipelines.
• Vision-language model security — technical framework for adversarial injection attacks targeting vision-language model image analysis pipelines, applicable to Palantir Foundry public health AI, HealthMap AI, and BlueDot AI VLM architectures processing outbreak map visualizations and genomic epidemiology display images.
• Free tier — 10 scans/day, no card required — start scanning public health surveillance AI inputs at development volumes; threshold 45 for pathogen biocontainment AI means every scan at a BSL-3 or BSL-4 facility monitoring boundary matters.
• Prompt injection scanner for healthcare AI — related adversarial attack surface covering clinical AI platforms with EHR, diagnostic imaging, and clinical decision support dimensions applicable to public health surveillance contexts where CDC BioSense and state health department AI intersect with hospital EHR syndromic surveillance data pipelines.
• PDF prompt injection detection — related adversarial attack surface covering AI pipelines processing PDF document images, applicable to MMWR outbreak report PDF page images, WHO IHR 2005 notification form PDF documents, and CDC FSAP inspection report PDF documents that generate page images for public health AI analysis.
• Multimodal AI security checklist — comprehensive security checklist for multimodal AI deployments applicable to CDC BioSense Platform AI, Palantir Foundry public health AI, and HealthMap AI implementations requiring systematic adversarial injection defense controls across outbreak map, genomic epidemiology, syndromic surveillance, and biocontainment AI image processing pipelines.