Survey blueprint: collapsing 42 categories into a fielding instrument

The wiki keeps 42 fine-grained categories (good for a buyer’s map). A CTO usage/evaluation survey can’t ask 42 questions — respondents fatigue and overlap confuses them. This page is the real instrument: ~16 questions, each mapping to one or more wiki categories, with the answer shortlist and design notes. Reuse the standard scale on every question:

Scale (multi-select per option): Interested · Considering/evaluating · Pilot/implementing · In production · Would recommend · Would not recommend.

Design rule: ask about what a fund buys as one decision. Where the wiki splits for precision (DSPM vs DLP; the five agent rows), the survey merges or gates with a screener.

Screener (route the respondent)

  • S1. Do staff use public/enterprise AI tools today? (yes → everyone gets the core)
  • S2. Do you do RAG over internal data? (gates the retrieval/data questions)
  • S3. Do you build/run agents that take actions? (gates the agent-security questions)
  • S4. Are you under SR 11-7 / formal model risk or heavy allocator DD? (gates governance depth)

Core questions (everyone)

#Question (category)Maps to wikiShortlist (answer options)
Q1Enterprise AI assistantenterprise-ai-assistantChatGPT Enterprise, Claude Enterprise, M365 Copilot, Gemini Enterprise, Amazon Q, Glean, Perplexity, Other
Q2AI gatewayai-gatewayPortkey (Palo Alto), LiteLLM, Kong, TrueFoundry, Cloudflare, F5, OpenRouter, Other
Q3AI access governance / shadow-AI + prompt DLPai-access-governance + dlpWitnessAI, Harmonic, Aurascape, Cyberhaven, Nightfall, Prompt Security (SentinelOne), Zscaler/Netskope-native, Other
Q4AI runtime security (firewall)ai-runtime-securityPrisma AIRS (Palo Alto), HiddenLayer, WitnessAI, Pillar, Cisco AI Defense, Lakera (Check Point), Lasso, Other
Q5LLM observability & evalllm-observabilityLangfuse (ClickHouse), LangSmith, Arize, Braintrust, Datadog, Helicone, Fiddler, Other
Q6AI governance / model riskai-governance-platform (+comms-surveillance sub-q)Credo AI, Holistic AI, ModelOp, IBM watsonx.governance, OneTrust, Vanta, Monitaur, Other
Q7AI-aware network security / SASEnetwork-security-sasePalo Alto, Zscaler, Netskope, Cloudflare, Cisco, Cato, Forcepoint, Other

Data & retrieval (gated by S2 = RAG)

#QuestionMaps toShortlist
Q8DSPM + DLP + data-access governance (one question, note overlap)dspm + dlp + data-access-governanceMicrosoft Purview, Varonis, Cyera, BigID, Securiti (Veeam), Sentra, Concentric, Immuta, Other
Q9Entitlement-aware RAGentitlement-aware-ragGlean, Microsoft Graph/Copilot, Knostic, Gemini Enterprise, Amazon Q, Other

Agent stack (gated by S3 = agents)

#QuestionMaps toShortlist
Q10AI-SPM / agent security (merges 2 wiki rows)ai-spm + agent-runtime-securityZenity, Noma, Prisma AIRS, Cranium, Lasso, Operant, Straiker, Other
Q11Agent authorization & tool/MCP access (merges 3 wiki rows)authorization-engine + mcp-gateway + tool-identity-integrationCerbos, OPA/Styra, Permit.io, Oso, Kong MCP, Solo.io agentgateway, Arcade, Composio, Other
Q12Non-human / agent identitynon-human-identityCyberArk (Palo Alto), Astrix (Cisco, pending), Aembit, Oasis, Token, Natoma (Snowflake), Entro, Clutch, Other

Foundation & governance (gated / as relevant)

#QuestionMaps toShortlist
Q13Identity & IGAidentity-access + identity-governanceEntra, Okta, Ping, SailPoint, Saviynt, Veza, ConductorOne, Other
Q14AI SOC / security automationai-soc-analystsProphet, Dropzone, 7AI, Radiant, Simbian, CrowdStrike Charlotte, Palo Alto AgentiX, Torq, Other
Q15Comms surveillance (AI-prompt capture)comms-surveillanceBehavox, SteelEye, NICE Actimize, Theta Lake, Shield, Relativity Trace, Other
Q16Enterprise browser / extensionenterprise-browser + browser-security-extensionIsland, Prisma Access Browser, Menlo, Seraphic, Chrome Enterprise, Edge for Business, LayerX, Other

Categories intentionally NOT given their own question

Folded into the above or out of scope for a usage survey:

  • Foundation incumbents (EDR, SIEM, secrets, SSPM, supply-chain, vendor-risk, GRC, anti-deepfake) — mostly “already owned”; ask only if the survey’s scope includes the general security estate.
  • AI red-teaming (ai-red-teaming) — fold into Q4/Q5 unless surveying AI builders; it’s a low-fit, developer-tool category for most funds.
  • Retrieval infra (content-sources, vector-retrieval) — plumbing, not a governance choice.
  • Process rows (trust zones, risk tiers, promotion gates, HITL, AUP) — better as maturity questions (“do you have a written AUP? risk-tiering? trust-zone design?”) than vendor questions.

Cross-cutting design notes

  • Label acquired options with their parent — many answer options are now product lines, not companies (see ai-security-m-and-a-map). A bare “Lakera” or “Portkey” dates the survey.
  • Expect double-counting where one vendor answers several questions (Palo Alto, Microsoft, Cyberhaven, WitnessAI, Lasso, Glean) — reconcile in analysis, not by forcing single-select.
  • Overlap traps to pre-empt: DSPM↔DLP↔access-governance (Q8); runtime↔red-team (Q4/Q5); the five agent rows (Q10/Q11/Q12); assistant↔entitlement-RAG (Q1/Q9).

History

  • [2026-06-28] Created in Phase 4 per taxonomy-gaps Q4 — the fielding instrument behind the 42-category wiki.