ConsultingWorkbench-backed AI security engagements — map, attack, defend, and prove your AI systems.
Scope a Review

Research Program

Continuous intelligence · Live signal monitoring

Research that sharpens AI security engagements.

Our research program tracks job-market signals, practitioner evidence, adversary frameworks, vulnerability disclosures, and public-source intelligence so our consulting engagements start from evidence rather than opinion.

When independent sources describe the same structural gap from different angles, convergence becomes the argument. That convergence feeds the annual report, named findings, assessment domains, engagement workshops, and every public claim in the citation library.

Signal Sources

Job Market

Ch 01

Practitioner Survey

Ch 02

Academic Velocity

Ch 03

Builder Ecosystem

Ch 04

Media Coverage

Ch 05

Threat Disclosure

Ch 06

ATLAS / Adversary

Ch 07

Citation Library

Ch 08

Convergence

Engine

The argument.

Directional
confidence

Outputs

Signal Convergence Score

92%

Directional confidence

Report

Findings

Methodology

Assessment domains

Live stack

8 signal channels

ATLAS / benchmark

170 techniques · 85+ signals

How this supports clients

Research supports the consulting method.

Better scoping for AI product security reviews
Stronger role and ownership models
Better framework mapping for governance evidence
More precise red-team scenarios
More credible executive and buyer-facing claims

Intelligence Channels

Eight sources. One thesis. Zero redundancy.

Each channel is independently collected and classified against the same AI security taxonomy. No single source is authoritative - the value is in triangulation. A finding that appears across job postings, papers, builder activity, media coverage, and vulnerability feeds is no longer a hunch; it is signal convergence.

01

Research Channels

Live Monitoring

8 independent monitoring channels - academic velocity, open-source velocity, press coverage, threat disclosure, concept maturity, MITRE ATLAS, controls crosswalk, and convergence.

ReportFindingsAssessment domains
Open channel

02

Annual Report

Flagship Publication

The State of AI Security Engineering 2026 distills labor-market, survey, and external signal data into named findings and role archetypes.

Executive briefingsSponsorsPublic research
Open channel

03

Intelligence

Synthesized Analysis

Named findings and market intelligence synthesized from job corpus analysis, practitioner surveys, and signal convergence.

FindingsWorkshopsConsulting
Open channel

04

Citation Library

Verified Sources

Verified citations with quotes, statistics, source provenance, dates, and claim traces.

ReportPublic claimsEvidence packs
Open channel

05

Threat Disclosure

Threat Signals

CVE, NVD, GHSA, OSV, and CISA KEV records classified for AI/ML relevance and disclosure lag.

Red teamDetectionAdvisories
Open channel

06

Convergence

Cross-Source Synthesis

Composite scoring where independent channels agree. The argument is in the convergence, not any single layer.

FindingsConfidenceRoadmap
Open channel

07

Academic Velocity

Research Frontier

Paper velocity, bucket-share composition, and term acceleration across classified AI security research papers.

TaxonomyFuture domainsField guide
Open channel

08

MITRE ATLAS Navigator

Adversary Framework

Tactics, techniques, mitigations, and case studies organized for practitioner use and cross-referenced against report findings.

LabsRed teamDetection
Open channel

Flagship Publication

The annual report is the distilled output of the research program.

2026 Edition

The State of AI Security Engineering Report

aisecurity.llc

The State of

AI Security Engineering

2026 Report

signal convergence

The State of AI Security Engineering 2026 is not a vendor survey or a trend roundup. It is a structured analysis of the AI security labor market: who employers are hiring, what skills they require, what gaps persist, and what the signal layers say about where the field is heading.

300K+ job descriptions. Practitioner survey data from four role cohorts. External corroboration across arXiv, GitHub, media, and vulnerability feeds. 15 flagship findings. 9 role archetypes.

Report scope

  • 300K+ AI security job descriptions analyzed
  • 4 practitioner survey cohorts
  • 8 external signal layers cross-corroborated
  • 15 named findings with evidence chains
  • 9 role archetypes mapped to skill clusters
  • MITRE ATLAS and OWASP LLM Top 10 crosswalk

Evidence spine

Built for briefing, not browsing.

The report sits on top of the intelligence stack, not beside it. Channels feed findings, findings feed the report, and the citation library keeps every public claim tied to provenance and date.

Executive briefings

Board-ready summaries with claim-to-control traceability.

Sponsor-safe public research

Published findings sponsors can reference without restriction.

Findings traceability

Every finding links back to channel evidence and source citations.

Claim-ready citations

Verified quotes and statistics with dates and provenance.

Channels
Findings
Report
Citations

Named Findings

Intelligence you can act on, not observations you already knew.

Methodology

How the research program works.

Data collection

  • Job description corpus: public ATS postings, classified by role and skill signal
  • Practitioner surveys: structured instruments across four cohorts
  • arXiv: metadata slices, paper velocity, and term acceleration
  • GHArchive: open-source activity across AI security repos
  • Vulnerability feeds: NVD, GitHub Advisory, OSV, CISA KEV, filtered for AI/ML relevance
  • Media/news coverage: public narrative signal and concept maturity

Taxonomy & classification

  • Unified AI security taxonomy across all channels
  • MITRE ATLAS crosswalk for adversarial technique alignment
  • OWASP LLM Top 10 mapping for product security coverage
  • NIST AI RMF alignment for governance and compliance signals
  • Citation verification with source, date, quote, and provenance
  • Convergence scoring across independent sources

Operating Outputs

Research that feeds engagements, not just PDFs.

Assessment Domains

The diagnostic scorecard and assessment domains are grounded in the research taxonomy.

Take diagnostic

Engagement workshops

Working sessions use report findings and field-guide controls to produce artifacts.

View workshops

Public proof surfaces

Scenario catalogs translate adversary and failure patterns into public demos and evidence-navigation surfaces.

Explore labs

Workbench-backed assessments

Assessments and evidence packs use the same language, controls, and citation base.

Scope a Review

Research Program

Use the signal stack behind the report.

Explore the channels, read the findings, or bring the research into your AI security program through diagnostic scorecards, workshops, and Workbench-backed assessment work.