AI Governance & Security Vendor Wiki

6 min read

SplxAI

Now SPLX, a Zscaler company. Acquired by zscaler — announced 2025-11-03, closed in Zscaler’s Q1 FY2026. Primary category: ai-red-teaming; also ai-runtime-security.

One-liner — Automated, continuous AI red-teaming (“Probe”) plus runtime guardrails and prompt-hardening for conversational and agentic AI apps — now owned by Zscaler.

What it does

SplxAI attacks your own AI applications before adversaries do. Its flagship product, Probe, runs thousands of automated adversarial simulations (the company cites 5,000+ attack scenarios) against chatbots, copilots, RAG pipelines, and agentic workflows — probing for prompt injection, jailbreaks, data leakage, hallucinations, toxic output, and off-topic/PII exposure across text, image, and voice. It then triages findings and recommends remediations (including prompt hardening — automatically tightening system prompts). Beyond pre-deployment testing, SPLX added runtime guardrails plus AI asset discovery and governance/compliance reporting, pushing it from a point red-teaming tool toward a fuller AI-security lifecycle platform. The team also open-sourced Agentic Radar, a static-analysis tool that maps dependencies in agentic workflows to surface missing controls.

The pitch is economic as much as technical: automated red-teaming is positioned as far cheaper and more repeatable than manual pentests (the company claims manual evaluation is ~5x more expensive — a vendor marketing figure).

Where it sits in the stack

  • Primary: ai-red-teaming (model/prompt layer) — the proactive, pre-deployment leg: find weaknesses in your AI app before go-live and on every change.
  • Secondary: ai-runtime-security (AI firewall) — runtime guardrails/threat inspection that block malicious prompts and unsafe responses in production.

Risk coverage. Red-teaming is overwhelmingly about untrusted input — stress-testing how the app handles hostile prompts/content. The runtime guardrails extend coverage toward sensitive-data exposure and exfiltration (blocking data leakage and unsafe tool/response paths). Net: it spans input screening, data protection and outbound controls, but its center of gravity is untrusted-input/prompt-injection. Trust zone: wherever LLM-facing apps live (typically yellow/green app tiers).

Deployment & architecture

  • SaaS platform with API access, so Probe scans can be wired into CI/CD and dev workflows (test the app on every release).
  • Runtime guardrails operate as an inline inspection/guardrail layer for production traffic (threat inspection + prompt hardening).
  • Connects to the AI apps/agents under test; post-acquisition the roadmap is integration into the Zscaler Zero Trust Exchange (AI asset discovery → red teaming → runtime guardrails → governance, “development through deployment”).
  • Open-source adjunct: Agentic Radar (static analysis of agent workflows).

Positioning & differentiators

SplxAI is known for automated, continuous red-teaming (not a one-off manual engagement) with a research-driven, continually-updated attack database and multi-modal coverage (text/image/voice). Founders and team came out of AI red-teaming at Cisco, Zscaler, and Wiz, with CTF wins (Wiz, Black Hat) as credibility.

Versus neighbors:

  • mindgard — closest pure-play automated AI red-teaming peer (research/academia roots); both target offensive testing of AI apps.
  • promptfoo — open-source/developer-first eval + red-teaming (acq. by OpenAI); SplxAI is the enterprise SaaS, governance-flavored counterpart.
  • enkrypt-ai — overlapping red-team + guardrails combo; both straddle testing and runtime.
  • lakera — strong on runtime prompt-injection guardrails (and red-teaming via Gandalf heritage); Lakera is more runtime-firewall-first, SplxAI more red-team-first.
  • pillar-security — runtime + posture for AI apps; overlaps on the runtime side.
  • hiddenlayer — model-centric security (model scanning + detection-response) and red-teaming; different emphasis (model artifacts vs app behavior).
  • prisma-airs, witnessai — runtime AI firewall / access-governance incumbents; SplxAI’s guardrails compete at the edges but its differentiator is the offensive testing side.

The Zscaler acquisition mirrors the broader consolidation: network/SSE incumbents buying AI-security capability to bolt onto existing inline enforcement points.

Ownership, funding & M&A

  • Funding (pre-acquisition): $7M seed, announced 2025-03-26, led by LAUNCHub Ventures, with Rain Capital, Inovo, Runtime Ventures, DNV Ventures, and South Central Ventures. Brief’s “led by LAUNCHub” — confirmed. No separate later priced round found before acquisition.
  • M&A — CONFIRMED: Acquired by Zscaler (NASDAQ: ZS). Announced 2025-11-03 (Zscaler press release + SPLX blog). Closed during Zscaler’s Q1 fiscal 2026; Zscaler disclosed Red Canary + SPLXAI together for an aggregate $692.0M — the SPLX-specific price is not separately disclosed (Red Canary was the much larger component). Ownership confidence: high.
  • The research brief assumed “independent”; that is now outdated (soft/scope, not a hard contradiction — the brief predates or missed the deal). Ownership reset to subsidiary (Zscaler).

Note: founding year — the seed announcement says founded 2023; SiliconANGLE says the product “launched August 2024.” Read as entity formed 2023, product GA mid-2024. Soft, non-blocking.

CTO / hedge-fund lens

Day-2, low direct fit for most hedge funds. SplxAI/Probe is a tool you need when you are building and shipping your own LLM/agent applications and want to test them adversarially — a developer/AppSec capability. A typical 50-person fund consuming ChatGPT Enterprise or Copilot does not red-team its own models and gets little from this. It becomes relevant only if the fund builds customer- or analyst-facing AI agents that handle sensitive data, in which case automated red-teaming + runtime guardrails are a sensible Day-2 control once the basics (access governance, DLP, an AI gateway) are in place.

Model-risk angle: the governance/compliance reporting can feed an SR 11-7-style model-risk file (evidence of adversarial testing), but SplxAI is a testing tool, not a governance platform — pair it with an ai-governance-platform vendor for that.

Practical note: because it is now part of Zscaler, shops that already run Zscaler SSE may eventually get this capability as a platform add-on rather than a standalone buy — worth tracking how the standalone product and pricing survive integration.

Competitors / alternatives

mindgard, enkrypt-ai, lakera, pillar-security, promptfoo, hiddenlayer, prisma-airs, witnessai

Open questions / to verify

  • SPLX-specific acquisition price (bundled with Red Canary at $692M aggregate; not broken out).
  • Whether Probe/runtime guardrails remain available as a standalone product, or only via the Zscaler Zero Trust Exchange, and any pricing/packaging changes post-integration.
  • Exact close date within Zscaler Q1 FY2026; precise founded month vs product GA.
  • CTO co-founder Ante Gojsalić — confirm current role post-acquisition.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; established $7M seed (2025-03-26, led by LAUNCHub Ventures + Rain/Inovo/Runtime/DNV/South Central), founded 2023 (Croatian roots, Delaware/NYC US), Probe automated red-teaming + runtime guardrails (SaaS/API). CONFIRMED acquisition by Zscaler (announced 2025-11-03, closed Q1 FY2026; bundled $692M with Red Canary, SPLX portion undisclosed) — ownership changed from independent→subsidiary, confidence high. Set hedge_fund_fit low (build-your-own-AI tool). Cached 4 sources.

Graph View

Backlinks