Securing AI Market Maturity Map

Marks after a name: ◌ innovating · ⚒ hiring · $ raised in the last year · "(was …)" moved stage since the previous map. Each mark belongs to the name before it.

Executing (6) — the largest vendors: scale of 75 and over

HiddenLayer, Zenity ◌, WitnessAI, Oasis Security, Noma Security ◌, Airia $

Scaling (15) — scale from 50 to 75

Straiker ◌ $, Quilr, Hush Security ◌ $, Cranium AI, Aurascape $, Rein Security $, Lasso Security ◌ $, Harmonic Security ⚒ $, Geordie AI, Singulr AI, Aembit ◌, Clutch Security, AIR Security $, Arthur AI, Mindgard $

Building (11) — scale from 25 to 50

Token Security ◌ ⚒, NeuralTrust ◌ ⚒ $, Operant AI ◌, Vijil ◌ $, Knostic ◌, DeepKeep ◌, Pillar Security, Adversa AI ◌, Giskard, MintMCP ⚒ $, Lumeus.ai

Emerging (7) — the smallest vendors: scale under 25

Guardrails AI $, Lunar.dev, Enkrypt AI ◌, Repello AI, Highflame, Zaun, Optimus Labs

Accelerating (2) — momentum of 67 and over

Zenity, Aembit

Growing (36) — momentum from 33 to 67

Straiker, MintMCP, Vijil, AIR Security, Hush Security, NeuralTrust, Aurascape, HiddenLayer, Knostic, Token Security, DeepKeep, Harmonic Security, Rein Security, Lasso Security, Noma Security, Adversa AI, Operant AI, Giskard, Enkrypt AI, Pillar Security, Mindgard, Geordie AI, WitnessAI, Zaun, Oasis Security, Cranium AI, Guardrails AI, Airia, Arthur AI, Quilr, Highflame, Clutch Security, Singulr AI, Repello AI, Optimus Labs, Lunar.dev

Holding (1) — momentum under 33

Lumeus.ai

Innovating (14) — top third by launches, corroborated launches, research posts and engineering hiring; a marker across every tier

Zenity, Straiker, Noma Security, Aembit, Hush Security, Lasso Security, Token Security, NeuralTrust, Vijil, Knostic, Operant AI, DeepKeep, Adversa AI, Enkrypt AI

Hiring (4) — top third by open roles per 100 staff, with at least 3 roles open; a marker across every tier

MintMCP (9 open roles, 60 per 100 staff), NeuralTrust (6 open roles, 17 per 100 staff), Harmonic Security (9 open roles, 10 per 100 staff), Token Security (5 open roles, 9 per 100 staff)

Raised in the last year (13) — a round dated inside the last 365 days, from the vendor's own post, a matched news event or the Crunchbase news list

MintMCP (round not stated, 2026-10), Lasso Security (Seed, 2026-10), Harmonic Security (round not stated, 2026-10), Rein Security (Series A, 2026-10), Guardrails AI (Seed, 2026-09), AIR Security (Seed, 2026-09), Mindgard (Series A, 2026-08), Hush Security (Series A, 2026-07), Straiker (Series A, 2026-06), NeuralTrust (Seed, 2026-06), Aurascape (Series A, 2026-04), Vijil (Series A, 2025-11), Airia (Seed, 2025-11)

Acquired (5) — on the record, not on the map

Promptfoo — by OpenAI, 2026-03-09; Katanemo — by DigitalOcean, 2026-04-01; Natoma — by Snowflake, 2026-05-27; Portkey — by Palo Alto Networks, 2026-05-29; TrojAI — by A10 Networks, 2026-06-15

Large incumbents (9) — in this market, but too large to compare on size; listed, not on the map

Check Point; Cisco; Cloudflare; CrowdStrike; F5; Kong; Palo Alto Networks; SentinelOne; Zscaler

• Zenity: Name collides with the GTK 'zenity' dialog tool; use the search_name for news queries.
• HiddenLayer: Broader than app/agent runtime; model scanning and supply chain are a large part of the product.
• WitnessAI: Weighted toward employee AI-use governance and policy rather than application-level runtime defence.
• Aembit: Workload / non-human IAM vendor first; AI-agent identity is an extension of the same platform rather than its origin.
• Hush Security: Launched in 2025 as secretless machine access (vault replacement); the AI-agent identity positioning came with the 2026 Series A. Partly general non-human identity.
• Oasis Security: General non-human identity management vendor; agentic access control is a 2025-26 extension of the NHI platform.
• Aurascape: Core product is employee AI-use control (shadow AI, data leakage via SaaS AI tools), closer to SSE/CASB than to LLM app or agent runtime security.
• Rein Security: New York and Tel Aviv; founded 2024 by Matan Bar-Efrat (CEO) and Netanel Rubin (CTO). Runtime security for AI agents at the point of action, sidecar rather than gateway. 'Rein' alone is a common word; keep 'Rein Security'.
• Airia: Broad enterprise AI platform (agent builder, orchestration, gateway); security and governance is one module, not the whole product.
• AIR Security: The company writes its name 'AIR' alone on most of its site; 'AIR' as a keyword will collide with ordinary text, so match on 'AIR Security' or the domain.
• Harmonic Security: 'Harmonic' alone collides with Harmonic Inc (video), Harmonic (math AI lab) and others; keep the compound.
• Token Security: Machine-first / non-human identity vendor; AI agents are one identity class it covers, not the core pitch. 'Token' alone collides with crypto and general usage.
• Cranium AI: Weighted toward AI inventory, governance and compliance (AI-SPM) rather than runtime or red teaming; the Aiceberg deal adds agent runtime.
• Geordie AI: 'Geordie' is a common word (Newcastle demonym); always search with 'Geordie AI'.
• Quilr: Launched in Apr 2025 as a human-risk / 'service-as-software' platform to prevent employee-caused breaches; the agent-governance positioning is from 2026. Small, seed-stage, with vendor-fit still shifting.
• Knostic: Endpoint and coding-agent focus rather than LLM application runtime protection.
• Clutch Security: General non-human identity and secrets security vendor; AI agents were added to the pitch in 2025. 'Clutch' alone collides with the automotive and sports senses.
• Arthur AI: Started as an ML model-monitoring vendor; the agent-security framing dates from the Dec 2025 ADG launch, so older coverage is about observability, not security.
• Adversa AI: Homepage in Oct 2026 pitches runtime controls for coding agents (Claude Code, Copilot, Cursor), a pivot from the red-teaming business; small team, funding undisclosed.
• Giskard: LinkedIn slug came from the homepage JSON-LD sameAs, not a visible footer link. 'Giskard' is also an Asimov character name; search_name may pull fiction references.
• Enkrypt AI: Thinly funded ($2.35M disclosed) relative to the rest of the group.
• Guardrails AI: 'Guardrails' is a generic term across the whole category; keep the compound 'Guardrails AI'. The only X account on the site is the Snowglobe product handle, not a company handle.
• Zaun: Moved from AI in the SOC (market 2 on bugfixing, marked pivoted): now security for AI. Product: Reagent, controls to govern and secure AI use and agents. 'Zaun' alone collides with the German word and the Arcane/League of Legends city; search as 'Zaun AI'. The site's footer links linkedin.com/company/zaun-ai, but company/zaun is the page whose profile reads website zaun.ai (bugfixing read, 26 Aug 2026).
• Lunar.dev: Began as an API consumption and rate-limit manager; security value rests on MCPX policy enforcement and auth, not a full security stack.
• Lumeus.ai: Small pivot-stage company (about 16 staff); website unreachable at time of check and status conflicting across directories.
• Optimus Labs: Secures the AI agent supply chain: finds the SKILLs, MCP servers, plugins and agent harnesses employees run, scores them against the OWASP Top 10 for Agentic Applications, and enforces policy at runtime. Legal name Optimus AI, Inc. 'Optimus' alone collides with Tesla's robot; search as 'Optimus Labs'. HQ country from the US corporate name; founding year not found.
• Promptfoo: Acquired by OpenAI (2026-03-09). Collected for the acquirer's news about the product.
• Katanemo: Acquired by DigitalOcean (2026-04-01). Collected for the acquirer's news about the product.
• Natoma: Acquired by Snowflake (2026-05-27). Collected for the acquirer's news about the product.
• Portkey: Acquired by Palo Alto Networks (2026-05-29). Collected for the acquirer's news about the product.
• TrojAI: Acquired by A10 Networks (2026-06-15). Collected for the acquirer's news about the product.
• Check Point: Lakera keeps its own site and LinkedIn (linkedin.com/company/lakeraai); the Check Point company page will carry mostly non-AI news.
• Cisco: 'AI Defense' without the Cisco prefix is a generic phrase; keep the full product name.
• Cloudflare: 'AI Gateway' and 'Firewall for AI' are generic without the Cloudflare prefix; the Firewall for AI name is being retired.
• CrowdStrike: 'Pangea' alone is a generic word (geology, Pangea Money Transfer); keep it only with context or drop it.
• F5: 'AI Guardrails' without the F5 prefix is the generic category term.
• Kong: 'Kong' alone is hopeless (King Kong, Hong Kong); only the product name is usable.
• Palo Alto Networks: 'Protect AI' as a keyword can match the generic phrase 'protect AI'; prefer exact-case matching.
• SentinelOne: 'Prompt Security' is also a generic phrase ('prompt security' in articles about prompt injection); require capitalised exact match. X handle belongs to the Prompt Security brand, LinkedIn is linkedin.com/company/promptsec.
• Zscaler: 'AI Guard' is shared with other vendors' products (Meta Prompt Guard, Lakera Guard); keep the Zscaler prefix.

Each input is a percentile rank among the rated vendors; an axis is the weighted mean of its inputs. On the map, the angle is the stage, by scale: Emerging under 25, Building from 25, Scaling from 50, Executing from 75, left to right. The distance from the base is momentum: holding under 33, growing from 33, accelerating from 67. A vendor moves right as it grows and outward as it speeds up.