SplxAI

Researched 2026-06-28; re-verified 2026-07-27 against Zscaler’s SEC filings. Now SPLX, a Zscaler company — acquired by zscaler, closed 2025-10-31 for $40.6M cash, publicly announced three days later on 2025-11-03. splx.ai is still live (HTTP 200, no redirect) under the SPLX brand and banners “SPLX is now part of Zscaler”; Probe still ships as a named product. Primary category: ai-red-teaming; also ai-runtime-security. No longer an independent company — shop for this inside a Zscaler platform conversation.

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.

Positioning has changed: this is now a platform feature, not a standalone bet. Since 2025-10-31 SPLX is a Zscaler product line, so the honest comparison is no longer “SPLX vs other red-teaming startups” but “Zscaler’s AI-security stack vs the other SSE/platform stacks.” That reframes almost every row below: the pure-plays are now the independent alternatives to a captive Zscaler capability, and the incumbent-owned tools are the true head-to-head competitors.

Versus neighbors:

• mindgard — closest pure-play automated AI red-teaming peer (research/academia roots); both target offensive testing of AI apps. Now one of the few remaining independent options in this niche if you don’t want to buy into a platform.
• promptfoo — open-source/developer-first eval + red-teaming (acq. by OpenAI); SPLX is the enterprise SaaS, governance-flavored counterpart. Both are now incumbent-owned.
• 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, SPLX more red-team-first. Now Check Point-owned, making this a Zscaler vs Check Point comparison rather than a startup bake-off.
• 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; SPLX’s guardrails compete at the edges but its differentiator is the offensive testing side. Prisma AIRS is the direct Palo Alto vs Zscaler platform rival.

The Zscaler acquisition mirrors the broader consolidation: network/SSE incumbents buying AI-security capability to bolt onto existing inline enforcement points. The AI red-teaming category has now been almost entirely absorbed this way — Lakera→Check Point, promptfoo→OpenAI, CalypsoAI→F5, SPLX→Zscaler — which means the practical selection question for a buyer is increasingly which platform vendor you already run, not which red-teaming engine tests best. The $40.6M price tag is worth noting for what it signals: a small tuck-in acqui-hire-plus-technology deal (about a third of the consideration was retention equity), not a category-defining purchase, so expect SPLX to show up as a Zscaler feature rather than a preserved standalone franchise.

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, dates and price corrected 2026-07-27. Acquired by Zscaler (NASDAQ: ZS). Closed 2025-10-31; publicly announced 2025-11-03. Price: $40.6M total cash purchase consideration, plus $16.6M grant-date fair value of restricted shares to retained SPLX employees (service-conditioned, so expensed post-combination rather than counted as purchase price) and 50,180 deferred shares for key-employee re-vesting. Source: Zscaler’s Form 10-Q for the quarter ended 2025-10-31 (filed 2025-11-25), Note 6 Business Combinations: “On October 31, 2025, we acquired all the equity of SPLXAI Inc.” Ownership confidence: high.
• The deal closed before it was announced. Zscaler’s 2025-11-03 press release says it “today announced it has acquired” SPLX — it publicised a transaction that had already completed on the last day of Zscaler’s fiscal Q1. There was no publicly-announced signing date and never a pending window, which is why announced post-dates closed in the frontmatter. That ordering is deliberate and correct.
• Price correction. This page previously said the SPLX price was “not separately disclosed” and folded it into the $692.0M aggregate. That aggregate is real but covers two deals that closed in the same quarter — Red Canary (closed 2025-08-01, the much larger component) and SPLX. The 10-Q does break SPLX out separately, at $40.6M.
• 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 (closed 2025-10-31), shops that already run Zscaler SSE may get this capability as a platform add-on rather than a standalone buy. As of 2026-07-27 SPLX still sells under its own brand with its own Probe login, so a standalone evaluation is still possible — but at a $40.6M tuck-in price, roughly a third of it retention equity, this was bought as technology and a team to fold into the Zero Trust Exchange, not as a franchise to preserve. Budget on the assumption it becomes a Zscaler SKU, and if you are not a Zscaler shop, weigh mindgard (the remaining independent) or whichever platform you have already standardised on.

Competitors / alternatives

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

Open questions / to verify

• SPLX-specific acquisition price — RESOLVED 2026-07-27: $40.6M cash, per Zscaler’s 10-Q Note 6.
• Exact close date within Zscaler Q1 FY2026 — RESOLVED 2026-07-27: 2025-10-31.
• 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. As of 2026-07-27 splx.ai still markets Probe under the SPLX brand with its own login at probe.splx.ai, which suggests standalone availability survives for now — but that is inference from the marketing site, not a confirmed Zscaler packaging statement.
• Precise founded month vs product GA.
• CTO co-founder Ante Gojsalić — confirm current role post-acquisition.

Sources

• Zscaler Form 10-Q, quarter ended 2025-10-31 (filed 2025-11-25), Note 6 Business Combinations — fetched 2026-07-27 — supports: close date 2025-10-31, SPLX-specific price $40.6M cash, $16.6M retention RSUs, 50,180 deferred shares, $13.0M developed technology / $1.1M customer relationships; also separates Red Canary (closed 2025-08-01) from SPLX within the $692.0M Q1 FY2026 aggregate; confidence: high. Cached: raw/sources/2026-07-27--splxai--zscaler-10q-close-date-price.md
• Zscaler press: acquisition of SPLX — fetched 2026-06-28, re-checked 2026-07-27 — supports: public announcement date 2025-11-03, phrased “has acquired” (already-closed deal, no closing-conditions language, no terms disclosed), products/scope; confidence: high
• SplxAI blog: $7M seed round — fetched 2026-06-28 — supports: $7M seed led by LAUNCHub + investors, founded 2023, Delaware HQ, Probe/runtime; confidence: high
• SiliconANGLE: SplxAI $7M / automated red teaming — fetched 2026-06-28 — supports: Probe mechanics, SaaS+API deployment, founders ex-Cisco/Zscaler/Wiz, customers (KPMG, Infobip, Glean), Agentic Radar; confidence: med
• Vestbee: Croatian SplxAI closes $7M seed — fetched 2026-06-28 — supports: Croatian roots, founders Kamber/Gojsalić, seed details; confidence: med

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.
• [2026-07-27] URL audit + M&A re-verification against primary SEC filings. splx.ai live (HTTP 200, no redirect), title “SPLX | End-to-End Security for AI”, banner “SPLX is now part of Zscaler” — brand and Probe product retained, so website unchanged. Corrected two frontmatter errors found by checking Zscaler’s Form 10-Q for the quarter ended 2025-10-31 (Note 6): (1) closed was 2025-11-03, a duplicate of the announcement date with no supporting source — the real close date is 2025-10-31, three days before the public announcement, because Zscaler announced an already-completed deal; (2) price was null and the body claimed the SPLX figure was “not separately disclosed” inside the $692.0M Red Canary + SPLX aggregate — the 10-Q does break it out, at $40.6M cash (plus $16.6M service-conditioned retention RSUs and 50,180 deferred shares). ownership_state: acquired-closed confirmed correct and now genuinely sourced. Rewrote competitive positioning for a platform-owned product (category now almost fully consolidated: Lakera→Check Point, promptfoo→OpenAI, CalypsoAI→F5, SPLX→Zscaler; mindgard remains independent). Cached 1 new source.