Comprehensive Agentic AI Security Startup Landscape — Pool-Then-Filter Pass

Why this is a gap

The May 2026 seed-funding synthesis ranked 8 startups + 3 honorable mentions. But the methodology was search-then-include, not enumerate-then-filter: web searches surfaced names, and each was verified against primary sources. There was no fixed candidate denominator.

That means the synthesis answers “what are the largest verifiable seed rounds we surfaced?” but does not answer:

  1. What is the full population of agentic-AI-security startups operating in 2025–2026? (any stage, any funding status)
  2. What fraction are seed vs. Series A+ vs. self-funded vs. acquired?
  3. Are there startups with smaller-but-real seed rounds we missed because they ranked below the search results’ fold?
  4. Which RA planes / CMM domains have the most vs. fewest startup entrants? (the synthesis answered this for the 8-startup subset, not the full population)
  5. Vendor concentration vs. fragmentation — is the field consolidating (a few well-funded leaders) or proliferating (many small entrants)?

Known starting points (curated landscapes)

These are published market maps that should anchor the pool-then-filter pass. None alone is exhaustive, but the union is a strong candidate set.

SourceURL / pointerCoverageNotes
Insight Partners — AI Agents Security Market Mapinsightpartners.com/wp-content/uploads/2025/10/AI-Agents-Security-Market-Map-Blog-Hero-4-2048x1152.jpg (Oct 2025)Categorized landscape — Insight’s own framingAlready a triangulated source for the wiki; the market map is the visual companion to their thesis writing
Crunchbasesearch: category:ai-security + category:cybersecurity + founded_date >= 2023-01-01Funding-stage dataAuthoritative for round size + date but quality of agentic-AI-security tagging is uneven
PitchBooksearch: AI security / LLM security / agent security verticalsFunding-stage dataBetter category fidelity than Crunchbase but paywalled
Menlo Ventures — AI Security LandscapePeriodic blog posts (track via menlo.com/posts)VC-curatedMenlo invested in General Analysis; their landscape view will skew portfolio-positive
a16z — AI security piecesa16z.com/topic/ai/VC-curatedBacked Keycard; same caveat as Menlo
Forgepoint — AI security thesisforgepointcap.comVC-curatedBacked Capsule
Gartner — AI TRiSM Market Guide (annual)Gartner subscriptionVendor-list authoritativeLags 6–12 months behind seed-stage market
OWASP GenAI / AISVS contributor listowasp.orgPractitioner-active vendorsIndicates which vendors are investing in standards work, not which exist
LinkedIn — #agenticAIsecurity and #MCPsecurity company listingsLinkedIn searchSelf-reportedUseful for stealth / pre-funding companies that don’t yet appear in funding databases

Methodology proposal

A rigorous version of the May 2026 pass should:

  1. Build the union pool from the sources above. Target: 50–80 candidates flagged as agentic-AI-security or AI-security adjacent.
  2. Tag each candidate with: founding year, latest funding stage (pre-seed/seed/A/B+), latest round date+size, geographic HQ, primary RA plane, primary CMM domain, gateway-vs-instrumentation classification (where applicable), public/stealth status.
  3. Filter to in-scope: founded ≥ 2023-Q1 OR pivoted-to-agentic ≥ 2024-Q4. Drop pure-MLOps and pure-model-scanning vendors that don’t address agent runtime / authorization / egress.
  4. Bucket by funding stage — not just seed. The seed view is one slice; an A/B+ view tells a different story (incumbents like Lakera, Protect AI, HiddenLayer anchor that bucket).
  5. Re-map to RA + CMM with the full pool, not the seed-only subset. Likely shifts D6 (Data/Memory/RAG) and D8 (Supply Chain) findings.
  6. Identify acquired / shut-down / dormant entries. CART consolidation has already taken SplxAI (Zscaler) and Promptfoo (OpenAI); the landscape view should flag others.

Output target

A new seed synthesis-style page, but landscape-shaped not seed-shaped:

  • Full population table (50–80 rows, sortable: name, stage, RA plane, CMM domain, last round)
  • Stage distribution chart (in prose / Mermaid)
  • RA plane + CMM domain heatmap showing where the field is dense vs. sparse
  • Acquisitions / consolidation list with dates
  • D6 (Data/Memory/RAG) re-test — was the May 2026 “zero seed-stage agentic-specific entrants” finding a real gap, or an artifact of the smaller search?

Trigger to execute

  • When the cumulative noise of “wait, did we cover X?” or “is Y in the wiki?” reaches ~5 unanswered names
  • After Insight Partners or Menlo publishes a refreshed (2026-Q3+) market map
  • Before the next wave of CMM revisions, so the tooling-map columns (Standards / OSS / COTS) reflect a triangulated population not a search-surfaced subset
  • Approximately Q3 2026 if no other trigger fires first

Effort estimate

  • ~3 hours for the pool build (Crunchbase + PitchBook + market-map cross-reference)
  • ~2 hours for the per-candidate tagging
  • ~2 hours for the synthesis page
  • Total: ~1 work-day — significantly larger than the May 2026 2-round autoresearch pass