AI Governance & Security Vendor Wiki

8 min read

TrojAI

Re-verified 2026-07-27 against A10’s primary disclosures and EDGAR. The June 2026 acquisition is closed, not pending — A10’s own release says it “has acquired” TrojAI, with no closing conditions stated, and the brand has already been integrated as “TrojAI by A10 Networks.” Price never disclosed. Primary category: ai-runtime-security.

One-liner — AI-security platform pairing build-time red teaming (TrojAI Detect) with runtime threat protection (TrojAI Defend); acquired outright by network-infrastructure vendor A10 Networks (NYSE: ATEN) on 2026-06-15.

Categories — ai-runtime-security, ai-red-teaming

What it does

TrojAI secures enterprise AI across two layers. TrojAI Detect is automated red teaming — it probes models, applications, and agents for vulnerabilities at build time (jailbreaks, prompt injection, data leakage, harmful output), increasingly with agent-led multi-turn attack chains, and maps findings to OWASP / MITRE ATLAS / NIST. TrojAI Defend is a runtime AI firewall — inline inspection of prompts/responses to block prompt injection, sensitive-data leakage, and policy violations in production, including for AI coding agents (Claude Code, Codex). A newer Agent Runtime Intelligence capability (private preview as of early 2026) captures full agent execution traces for deeper runtime visibility.

Where it sits in the stack

Spans ai-runtime-security (Defend, the inline firewall — its primary slot) and ai-red-teaming (Detect, build-time testing). In lethal-trifecta terms it addresses untrusted-input (catching prompt-injection/jailbreaks both pre-prod and inline) and sensitive-data (blocking data leakage in responses). It lives at the model/prompt boundary between users/apps and the model, and is meant to sit in the yellow zone in front of model calls.

Deployment & architecture

API-based runtime inspection (Defend) that can be inserted in front of model endpoints / via an AI gateway; red-teaming (Detect) runs against model and app endpoints. Supports self-hosted / data-sovereign deployment — a point A10 emphasizes (“sovereign AI security”). Findings map to OWASP/MITRE/NIST; targets SIEM/SOC workflows and MSSP delivery. Post-acquisition, pairs with A10’s hardware-based AI firewall appliances.

Positioning & differentiators

One of the few players covering both build-time red teaming and runtime defense in one platform (vs. red-team-only mindgard/splxai or runtime-leaning prompt-security). Differentiators: agent-led/multi-turn red teaming, coding-agent runtime protection, and a sovereignty/self-hosted posture attractive to regulated and non-US buyers and MSSPs. Canada-rooted with Boston go-to-market. Nearest neighbors: hiddenlayer, prisma-airs, cisco-ai-defense, pillar-security, lakera, enkrypt-ai.

Ownership, funding & M&A

Acquired outright by A10 Networks (NYSE: ATEN) — announced and closed 2026-06-15. Re-verified 2026-07-27.

Why “closed” and not “pending.” A10’s own release reads: “A10 Networks (NYSE: ATEN) today announced that it has acquired TrojAI…” — completed-transaction grammar, with no definitive-agreement, closing-conditions, or expected-close language anywhere in the document. This is an ordinary simultaneous sign-and-close: a small private tuck-in needing no regulatory clearance. Two independent corroborations: the wire copy is identical across BusinessWire and StockTitan, and the brand integration is already live — troj.ai now ships a logo reading “TrojAI by A10 Networks,” which is post-close behaviour (pre-close, that degree of brand merger is precisely what gun-jumping rules forbid).

Price: undisclosed and not derivable. Terms were never released, and there is no filing to bound them against. A10 stated it “does not expect the acquisition to have a material impact on its financial results for fiscal year 2026.” Consistent with that, no 8-K was filed — Item 2.01 is only triggered by acquiring a significant amount of assets, so an immaterial deal creates no obligation — and EDGAR full-text search returns zero “TrojAI” hits across all A10 10-K/10-Q/8-K filings. The absence of a filing is what a closed-but-immaterial tuck-in looks like, not evidence against closing. A10’s first 10-Q covering the deal (quarter ended 2026-06-30) is due early August 2026 and is the first document that could carry a purchase-price allocation; verify_after: 2026-08-15 is queued against it.

Company. Founded 2019 (incorporated 2020) by James Stewart (CTO) and Stephen Goddard; CEO since 2024 is Lee Weiner (ex-Rapid7). HQ Saint John, New Brunswick, Canada (independently confirmed against TrojAI, Inc.’s SEC Form D business address), with a Boston office. Pre-acquisition funding ~$9.65M disclosed (incl. $5.75M seed Apr 2024 — Flying Fish Partners, Build Ventures, Techstars Ventures; plus NBIF). ownership_confidence: high.

What A10 gets — and why this is a diversification, not a bolt-on

A10 is a networking-infrastructure company: application delivery controllers, load balancers, DDoS mitigation, ~$250M revenue, historically sold to network and infrastructure teams. TrojAI moves it up the stack from the packet/flow layer to the model and prompt layer — a genuinely different discipline sharing almost no engineering substrate with ADC hardware, and sold to a different buyer (AppSec / AI platform teams rather than network ops). Read it as A10 buying a foothold in AI security rather than extending an existing line.

The stated thesis is “sovereign AI security”: pair A10’s on-box/hardware AI firewall with TrojAI’s software Detect + Defend so the entire pipeline can run inside a customer’s own boundary. That self-hosted posture is the most durable part of the rationale for regulated and non-US buyers — and it is the capability most worth watching, since it is also the one a new parent could dilute in favour of appliance attach.

CTO / hedge-fund lens

Day-1 for the runtime-firewall job if you expose LLMs/agents to users or untrusted content; the red-teaming half is a strong Day-2 assurance layer feeding promotion-gates. SR 11-7 / model-risk angle: Detect produces framework-mapped test evidence useful for model validation; Defend gives an enforcement + logging control. The A10 acquisition is a double-edged consideration for a hedge fund: it adds a public-company parent and appliance integration (good for procurement stability and on-prem/sovereign needs), but the product is now part of a networking vendor’s roadmap, so watch for repositioning. Self-hosted/sovereign deployment is a plus for funds wary of routing prompts through a vendor cloud. Fit medium — credible, but smaller and less proven at scale than prisma-airs or hiddenlayer; the new owner changes the buying calculus.

Competitors / alternatives

hiddenlayer, prisma-airs, cisco-ai-defense, pillar-security, lakera, enkrypt-ai, mindgard, splxai, prompt-security.

Open questions / to verify

  • Acquisition close date and integration/rebrand under A10. Resolved 2026-07-27 — closed 2026-06-15 (simultaneous sign-and-close); brand now “TrojAI by A10 Networks” on troj.ai.
  • Purchase price — never disclosed and no 8-K exists. Check A10’s 10-Q for the quarter ended 2026-06-30 (due early Aug 2026) for a business-combination note or cash-paid-for-acquisitions line; A10’s own materiality statement means it may never be broken out.
  • Whether TrojAI Defend will be bundled with A10’s firewall appliances vs. sold standalone, and whether the standalone SaaS survives.
  • Whether the newer Defend for MCP and Defend for Employees SKUs (live on the post-acquisition site) are generally available or still early — they postdate the last product-detail research pass.
  • Retention of Lee Weiner and the founding team under A10; no retention terms were announced.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; ownership changed independent → acquired (A10 Networks, announced 2026-06-15, terms undisclosed) — NOT in seed flags; set ownership_confidence high. Established founding 2019 (Stewart/Goddard), CEO Lee Weiner, HQ Saint John NB, ~$9.65M raised pre-deal. Documented Detect (red team) + Defend (runtime) split. hedge_fund_fit = medium.
  • [2026-07-27] URL audit + M&A re-verification against primary sources and EDGAR. acquired-closed CONFIRMED — no correction needed. The 2026-06-28 pass had set closed equal to announced with no separate citation (the defect shape found on natoma and symmetry-systems), so it was re-checked from scratch. It holds: A10’s release says it “has acquired” TrojAI with no closing conditions stated, the wire copy is identical across two independent retrievals, and troj.ai has already rebranded to “TrojAI by A10 Networks” — brand integration is post-close behaviour. Verified A10 filed no 8-K and that EDGAR full-text search returns zero “TrojAI” hits in any A10 periodic filing, consistent with A10’s own statement that the deal is immaterial to FY2026. Price remains undisclosed and undeterivable; left null. Confirmed HQ against TrojAI, Inc.’s SEC Form D. Confirmed A10 trades on NYSE (EDGAR exchanges: ["NYSE"]) — the page was already right. Set last_verified: 2026-07-27, verify_after: 2026-08-15 (A10’s Q2 10-Q, the first filing that could disclose price); added 4 sources: entries and tags. Added a section on why this is a diversification for a networking vendor; resolved the close-date open question and added price / SKU / retention ones. survey-v2.csv row corrected — it said “pending,” which was wrong.

Graph View

Backlinks