SPLX

• Zscaler could retire the standalone SPLX product and brand as it integrates the technology into the Zero Trust Exchange, ending independent availability for existing SPLX customers.
• Zscaler's acquisition announcement flagged retention of SPLX employees as a risk, so departures of the founders or red-team researchers could slow the roadmap after integration.
• Model providers such as OpenAI and Anthropic, and platform vendors, could ship native AI red teaming and guardrails for the agents built on their platforms, compressing the standalone market SPLX served.
• The named customers KPMG, Infobip, and Glean and the 127% quarterly-growth figure trace to a single 2025 funding announcement, so a buyer checking paying-customer depth could find less independent confirmation than the traction implies.
• SPLX's attack-simulation library is a maintained catalog rather than a demonstrated cross-customer data flywheel, so a funded rival could rebuild comparable red-teaming coverage.

SPLX treats the AI systems an enterprise builds and adopts as the asset under attack. Its platform and the acquiring vendor's own account frame the problem as a fast-expanding attack surface of models, agents, workflows, code repositories, and Model Context Protocol servers that traditional security tools do not inspect, with named failure modes including prompt injection, jailbreaks, and data leakage across text, images, and voice.

The buyer is the enterprise security, engineering, and risk team standing up AI in production. SPLX positions the pain as spanning the whole lifecycle, which is the shift-left framing the acquirer repeated when it described securing AI investments from development through deployment.

Independent reporting places the problem beyond vendor marketing. Trade coverage of the seed round and the acquisition describes the company as addressing the untested attack surface of enterprise AI applications and agents, which corroborates that the problem is real even though the quantified scale of the pain stays framed by the vendor and analysts rather than measured across buyers. [s2, s5, s7]

SPLX sells one platform across the AI security lifecycle rather than a single control. It discovers models, AI workflows, MCP servers, and guardrails into a complete AI bill of materials, runs automated red teaming that stress-tests those systems, inspects runtime traffic, and generates hardened system prompts to remediate the risks its testing finds.

The red-teaming engine is the center of gravity. The acquirer's account credits more than five thousand purpose-built and domain-specific attack simulations spanning development to production, and SPLX's own research arm published evaluations of frontier models including GPT-5, Claude Opus 4.1, and Grok 4, which both exercises and extends the attack library.

Open-source output gives the capability claims an outside check. SPLX released a tool that maps dependencies in agentic AI workflows using static code analysis, which independent press highlighted as part of its toolkit. That public tooling, plus the detailed platform pages, is the evidence a technical buyer can inspect without a sales conversation. [s3, s2, s1, s4]

SPLX competes as an end-to-end AI-security platform against rivals who each concentrate on one of its lines. Automated red-teaming specialists contest its testing engine, runtime-guardrail vendors contest its inline protection, and broader AI-security platforms pitch the same discovery-to-defense loop across models and agents.

Its visible differentiator was the connected path from discovery through red teaming to runtime remediation, sold as one platform a buyer would otherwise assemble from several vendors. The recurring public research on frontier models gave SPLX a red-team profile larger than its headcount, which functioned as both credibility and demand generation.

The risk is who owns the buyer. Model providers can test and guard the agents built on their own platforms, and platform vendors can bundle red teaming and guardrails into deals an enterprise already signs. The Zscaler acquisition is that exposure realized in the company's favor, a platform vendor bought the capability and made it part of its own suite. [s2, s3, s6]

SPLX's clearest traction is a set of named enterprise customers reported by independent press. Coverage of the seed round names KPMG, the telecommunications provider Infobip, and the generative AI search company Glean as users, and reports quarterly revenue growth of 127% since the platform launched in August 2024. Those names sit in independent reporting rather than only on the vendor's own pages.

The exit is consistent with the early traction. Zscaler acquired SPLX within about two and a half years of founding, which fits the customer names being real, though the deal price was not disclosed and does not verify paying-customer depth.

Depth behind the customer names stays undisclosed. The growth figure and the customer list trace to a single 2025 funding announcement, and the sources do not confirm paying-customer depth, contract size, or reference willingness, so the named traction is genuine but thinly corroborated beyond that announcement. [s4, s7, s6]

SPLX's founders paired offensive-security craft with a fast, completed exit. Co-founders Kristian Kamber, the chief executive, and Ante Gojsalic founded the company in Croatia in 2023 and sold it to Zscaler within about two and a half years, a completed in-domain acquisition that is the strongest public signal about the team.

The published research is the team's other durable signal. SPLX's researchers published red-team evaluations of frontier models, a body of public technical work rather than a single disclosure, which demonstrates the adversarial-AI expertise the commercial platform automates.

What the record does not show is a prior category-defining exit or sustained industry leadership before SPLX, which is the difference from the founders who earn the top mark. The build-and-sell outcome and the research record place the team above the plausible-background tier and level with the verifiable-build peers. [s5, s6, s1]

SPLX asks a buyer to let its software probe and inspect proprietary AI systems, so trust is load-bearing, and its strongest proof points are hands-on rather than documentary. The recurring public red-team research on named frontier models and the open-source workflow-analysis tool let a technical buyer judge the team's craft directly before committing.

Named enterprise adoption doubles as a trust signal. Independent reporting names KPMG, Infobip, and Glean as customers the company was onboarding, which signals to a prospect that security-conscious organizations engaged the product and lowers the perceived risk for the next buyer.

As an acquired company, SPLX's compliance and assurance posture now sits under Zscaler's program rather than standing alone. The reviewed sources document the product's capabilities and its enterprise users but not a standalone trust portal, so a procurement team would establish attestation and support terms through the acquirer rather than the former startup's own site. [s1, s4, s6]

Add analyzed competitors to compare them side by side with SPLX.

SPLX sold to the enterprise security, engineering, and risk team standing up AI in production, the owner accountable when a model, agent, or workflow is compromised. Its platform framed the target as the organization securing the whole AI lifecycle from development through deployment, a security-owner buyer rather than a developer or a line-of-business team.

The segment was broad by design, which was both the pitch and the exposure. SPLX spanned discovery, offensive testing, runtime protection, and governance in one platform, so it could enter through whichever pain a buyer felt first, at the price of defending an unusually broad claim for its stage at every edge of that span.

Named demand reached recognizable enterprises. Independent reporting places KPMG, the telecommunications provider Infobip, and the AI search company Glean as users, spanning professional services, telecommunications, and technology, which shows the platform addressed a serious enterprise buyer rather than a single vertical.

The product line was unusually complete for the company's stage. SPLX discovered models, AI workflows, MCP servers, and guardrails into a complete AI bill of materials, ran automated red teaming across text, images, and voice, monitored and blocked live inputs and outputs of AI systems in production, and generated hardened system prompts to remediate what its testing found.

The durable advantage sat in the attack corpus. The acquirer credited more than five thousand purpose-built, domain-specific attack simulations, SPLX's research arm published evaluations of frontier models such as GPT-5, Claude Opus 4.1, and Grok 4, and its product materials say hardened system prompts and targeted remediation steps are generated from red-teaming insights. Runtime traffic inspection ships as a separate platform function, and the reviewed sources do not state that the runtime guardrails consume the attack corpus.

The verifiable footprint backs the claim that the team can build. SPLX released an open-source tool that maps dependencies in agentic AI workflows through static code analysis, and its model-security research is public and inspectable. The unverified piece is the corpus size itself, a vendor-stated count with no third-party audit or accuracy benchmark in the reviewed sources.

SPLX ran a demo-led enterprise motion with a research-and-open-source top of funnel. The platform routed to a book-a-demo call to action rather than self-serve signup, which fits a security-team buyer and a negotiated deal, and the recurring public red-team research plus the open-source tool gave a prospect proof of craft before any sales conversation.

The traction it disclosed was concrete but narrow. Independent reporting names KPMG, Infobip, and Glean as customers and reports 127% quarterly revenue growth since the August 2024 launch, and the company reached those milestones on about nine million dollars raised across two rounds.

The exit is consistent with that traction. Zscaler acquired SPLX within about two and a half years, though the deal price was undisclosed and the depth of the customer relationships was not detailed beyond the funding announcement.

SPLX published no pricing in the reviewed sources. The platform converted to a demo request, the standard posture of a vendor selling negotiated enterprise contracts to security teams, which withholds the budget-anchoring signal some peers publish.

The charged unit is not stated. A platform spanning discovery, red teaming, runtime protection, and governance could meter on AI assets discovered, attack runs, protected traffic, or a flat platform subscription, and none of these is disclosed. The open-source tool and the public research were free top-of-funnel rather than a priced tier.

The inferable belief is that buyers paid for coverage of an expanding AI attack surface across the lifecycle rather than for a feature count, given the test-protect-govern framing. Confirming the unit and whether testing and runtime were bundled would have required the sales conversation the demo-only posture signaled.

The reviewed sources present SPLX as software the customer operates rather than a managed service. The AI-BOM, automated red teaming, runtime inspection, and remediation are functions a security team runs against its own AI systems, and nothing in the sources describes SPLX analysts running the testing or accepting accountability for the safety outcome on the customer's behalf.

The architecture was built to lower the operational lift. SPLX generated hardened system prompts and targeted remediation from its red-teaming insights without manual overhead, and the integrated loop was pitched as one platform spanning discovery, testing, remediation, and guardrails, so the customer did not stitch point products together.

Operational assurance now points toward the acquirer. The runtime guardrails sit in the live traffic path, where uptime and support terms matter, and given the acquisition and Zscaler's plan to integrate the guardrails into the Zero Trust Exchange, a buyer would increasingly evaluate those terms through the Zscaler relationship rather than the standalone startup, which a buyer would establish in a security review.

SPLX asked a buyer to let its software probe and inspect proprietary AI systems, so trust was load-bearing, and its strongest proof points were hands-on. The recurring public red-team research on named frontier models and the open-source workflow-analysis tool let a technical buyer judge the team's craft directly.

Named enterprise adoption reinforced that signal. Independent reporting names KPMG, Infobip, and Glean as customers the company was onboarding, which signals to a prospect that security-conscious organizations engaged the product and lowers the risk for the next buyer better than a self-asserted claim would.

SPLX displays ISO 27001 and SOC 2 Type II badges on its homepage and platform pages and maintains a trust center naming both attestations with the underlying materials behind a request-access gate, all re-verified live on 2026-07-15. The reviewed sources do not confirm the attestations' post-acquisition scope, so a procurement team would validate coverage through the Zscaler relationship as well as the request flow.

SPLX positioned itself as the platform that closes the loop across the AI security lifecycle rather than a point tool. It discovered AI assets into a bill of materials, tested them, protected them at runtime, and governed them, so the platform claim rested on owning the connection between offense and defense a buyer would otherwise integrate across several vendors.

The integration was the real differentiator. A customer assembling a red-teaming tool, a guardrail product, and an asset-discovery scanner from three vendors carries the integration burden, while SPLX argued its red-team findings fed remediation and prompt hardening within the same platform loop. The open-source workflow tool extended the ecosystem into agentic AI workflows.

The platform layer beneath SPLX is where the exposure lived, and the acquisition recast it. Rather than a model or cloud provider absorbing the capability natively, a zero-trust platform vendor bought SPLX and made it a dedicated AI-protection layer of its own exchange, so the independent platform became part of a larger one.

SPLX's credibility rested on offensive-security craft and a fast, completed exit. Co-founders Kristian Kamber, the chief executive, and Ante Gojsalic founded the company in Croatia in 2023, and it was acquired by a NASDAQ-listed security leader within about two and a half years, a completed in-domain outcome.

The published research is the team's strongest ongoing signal. SPLX publishes red-team evaluations of frontier models including GPT-5, Claude Opus 4.1, and Grok 4, a body of public technical work that demonstrates the adversarial-AI expertise the commercial platform automates.

What the record does not show is a prior category-defining exit before SPLX or sustained standards-body leadership. The build-and-sell result and the research record demonstrate execution and adversarial-AI expertise that a buyer can inspect directly.