NeuralTrust

• Large API and platform vendors are moving toward AI traffic management, so an enterprise could eventually get gateway-level LLM screening inside a contract it already signs and skip a standalone NeuralTrust purchase.
• NeuralTrust's gateway is open-source, which lowers switching cost and lets a larger vendor study and match the design, eroding the foothold before paid runtime and red-teaming revenue compounds.
• Even with a $20 million seed, NeuralTrust holds far less capital than the platform vendors and cloud providers entering AI-gateway security, so it must convert that raise into named enterprise references before a better-funded rival bundles a competing control.
• Named enterprise customers appear mainly in funding coverage rather than as referenceable deployments with public case studies, so buyers who require current speaking references could stall deals for a company this young.
• NeuralTrust's differentiation rests heavily on its jailbreak research, so a slowdown in that disclosure stream or a key researcher departure could remove the main reason a buyer chooses it over a bundled competitor.
• The founders show no prior security exit, so a buyer is betting that a research-strong team can also build the enterprise sales motion that turns a free gateway into recurring revenue.

NeuralTrust treats the live traffic flowing into an enterprise's GenAI applications and agents as something security teams cannot inspect or control with conventional tools. The company frames the buyer as the security team standing up LLM chatbots and agents inside regulated processes, where a prompt injection or a data leak reaches production systems. Its documentation positions the product as the runtime and red-team layer for that traffic.

The risk NeuralTrust points to is demonstrated, not hypothetical. Its own researchers built the Echo Chamber jailbreak, a multi-turn context-poisoning attack that The Hacker News reported bypassed GPT-4o and Gemini guardrails, which establishes that the safety filters enterprises rely on can be steered into producing disallowed output. A bank deploying a customer-facing chatbot is the concrete buyer this work speaks to. [s3, s8, s1]

The NeuralTrust platform spans a runtime layer, an agent-security layer, and an offensive testing layer rather than a single control. TrustGate is an open-source AI gateway that routes and screens LLM traffic and installs through a public self-hosted install command, and the company announced a Generative Application Firewall, while the cited pages ground runtime inspection and prompt-injection blocking in TrustGuard. Around the gateway sit prompt moderation, sensitive-data masking, bot detection, and agent-monitoring modules.

TrustTest is the offensive half, an automated red-teaming engine the company says carries tens of thousands of real-world adversarial attacks drawn from its own research. NeuralTrust folded the Echo Chamber jailbreak into both TrustGate and TrustTest so customers can simulate the multi-turn attack before it reaches production.

The research output demonstrates the capability the product sells. NeuralTrust researchers wrote Echo Chamber up as an arXiv preprint with a co-author from Universitat Pompeu Fabra in Barcelona and disclosed a further jailbreak against GPT-5 that Dark Reading reported, work that shows the team finds the attacks its firewall is built to stop. [s2, s7, s11, s16, s23]

NeuralTrust competes in AI gateway and runtime security against both independents and the platform vendors moving into the same slot. Lakera, which Check Point acquired, and Prompt Security cover overlapping runtime guarding and red teaming, while larger API and platform vendors move toward AI gateway offerings of their own. The category NeuralTrust sells into is one larger vendors are building toward.

The open-source gateway is NeuralTrust's adoption entry point and its structural weakness at once. Giving TrustGate away wins installs and a place in customer traffic, but a gateway is the layer a platform vendor can bundle into a deal an enterprise already signs, and an open-source one is straightforward for a larger player to match. Its durable separation is the research brand, which a bundled competitor cannot quickly reproduce. [s2, s8, s10]

Research is NeuralTrust's clearest go-to-market engine, and it generates attention rather than named revenue. The Echo Chamber and GPT-5 jailbreak findings drew coverage in The Hacker News and other outlets, and Gartner named NeuralTrust in guides for AI gateways and guardian agents, which builds inbound awareness and analyst validation. This is demand generation through research and analyst recognition rather than proof of paid deployments.

Verifiable customer proof is thinner than the recognition. A named testimonial from the Spanish bank ABANCA describes stress-testing a customer chatbot for a safe go-live, and the homepage logo wall shows ISDIN, Iberia, Sabadell, and Air Europa, but these appear as vendor-displayed logos and testimonials rather than independently reported references. A Gartner Peer Insights vendor page lists NeuralTrust in its AI Gateways market based on public information, a Gartner surface distinct from the company's own announcements of its Gartner recognition. NeuralTrust also won best cybersecurity startup at 4YFN during MWC 2026 and earned ISO 27001 certification.

The motion is early and direct. NeuralTrust routes prospects to a demo request and a free gateway download, leaning on its research and analyst profile to open enterprise conversations. A publicly referenceable paying enterprise would be the signal that this attention has converted into deployments. [s8, s6, s5, s12, s13, s15, s18, s20]

NeuralTrust's founders pair business-scaling experience with a strong research bench rather than prior security exits. Co-founder and CEO Joan Vendrell previously held roles at Mango, Amazon, and McKinsey, and co-founders Victor Garcia (engineering) and Alejandro Domingo (GTM) appear on the company team page, but the public record shows no prior security company built and sold. The leadership signal is commercial and operational, not a track record of shipping security products.

The research record is the team's strongest public asset. NeuralTrust researcher Ahmad Alobaid and colleagues wrote the Echo Chamber jailbreak up as an arXiv preprint, co-authored with Carlos Castillo of ICREA and Universitat Pompeu Fabra in Barcelona and with CEO Joan Vendrell among the authors, and the team disclosed a follow-on GPT-5 jailbreak reported by Dark Reading. A preprint is published research, not peer-reviewed work, but it is a sustained, externally checkable stream of in-domain adversarial output for a company this young.

The recognition gives that work outside validation. Independent outlets covered the Echo Chamber disclosure and Gartner cited NeuralTrust across multiple AI security guides, which separates a sustained research program from a single covered finding. [s8, s11, s5, s16, s17, s22, s23]

NeuralTrust's trust posture centers on a published certification and a flexible deployment model. The company states it is ISO 27001 certified, the recognized information-security management standard that a procurement team can ask to see, which is a stronger starting artifact than the same-asset peers that show no attestation. The announcement names Insight Assurance as the issuer, following an independent audit, and states that the certificate runs through 2029 with annual surveillance audits, so a buyer can check the issuer and the certificate's currency from the public page. For a vendor whose product inspects proprietary AI traffic, that certification addresses an early security-review question.

The company's trust portal at trust.neuraltrust.ai renders only a sign-in form, so the public verification path is the announcement rather than a browsable document set. A buyer who wants the certificate or supporting documents requests portal access rather than downloading them from a public page.

Deployment choice supports buyers who cannot send traffic to a vendor cloud. NeuralTrust documents SaaS, hybrid, and self-hosted modes built on a control-plane and data-plane split, and the open-source gateway can run inside the customer's own environment. A bank or healthcare buyer that needs sensitive prompts to stay in-house can run the data plane locally, which lowers the exposure objection that a runtime inspection product raises. [s5, s3, s12, s19]

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

NeuralTrust sells to the large, regulated enterprise standing up AI agents across many teams. The company describes a buyer where one team builds a customer-service agent on one model provider while another automates back-office work on a different one, leaving security teams unable to say how many agents run or what each may do. That fragmentation is the problem it scopes, and its named customers are global banks and airlines, with government customers reported but unnamed, rather than small teams.

The customer base is concentrated at the top of the market and tilted toward Europe. NeuralTrust reports that most of its customers are billion-dollar enterprises, the bulk of them based in Europe, and names Iberia, AirEuropa, Abanca, and Banc Sabadell among them. That profile fits a product whose self-hosted and air-gapped modes target defence, government, and the financial sector.

The European concentration is a deliberate foothold as much as a starting point. TechFundingNews reports that several of NeuralTrust's government and banking customers seek providers headquartered outside the United States as EU AI Act compliance requirements tighten, which gives a Barcelona vendor a procurement opening tied to where it is headquartered rather than to product features. The company states a goal of leading the European market by the end of 2026 before expanding internationally.

NeuralTrust's claimed advantage is owning the single point every agent call passes through. TrustGate is an agent gateway that brokers each model, tool, and protocol request, which the company frames as turning thousands of individual security decisions into one a customer controls and meters. It routes calls to any model, OpenAI, Anthropic, or self-hosted, and adds inline prompt inspection, PII redaction, and semantic caching at that chokepoint.

The runtime and offensive layers extend the gateway rather than wrap a single model. TrustGuard inspects every interaction and tracks conversation context across turns, catching a multi-turn attack where a jailbreak fails once and lands on a later attempt. Tech.eu describes the core platform as three products, TrustGate, TrustGuard, and TrustLens, while NeuralTrust separately markets TrustTest, a red-teaming engine that runs algorithmic adversarial probes against a customer's LLMs and agents across a taxonomy spanning prompt injection, indirect injection, and data leakage. TrustLens maps every agent in the enterprise and tracks its behaviour.

The research record is the externally checkable proof of the team's adversarial depth. NeuralTrust documented the Echo Chamber multi-turn jailbreak, reported by The Hacker News, and the Semantic Chaining multimodal attack, and Tech.eu reports both have been incorporated into the OWASP AI Security Project taxonomy. The durable engineering asset is real, though TrustGate routes to commercial or self-hosted models rather than a model NeuralTrust owns, and the flagship attacks are now published.

Research and analyst recognition are NeuralTrust's most visible demand surfaces. The Echo Chamber and follow-on jailbreak findings drew security-press coverage, and Gartner named the company a Representative Vendor in two AI-security Market Guides, third-party surfaces that build inbound awareness, though the company's lead sources and sales mix are not disclosed. A Gartner Peer Insights vendor page lists NeuralTrust in its AI Gateways market based on public information, a Gartner surface distinct from the company's own recognition announcements. The company pairs that with an open-source gateway download meant to seed adoption ahead of a paid upgrade.

The named-customer record is corroborated twice over. TechFundingNews independently names Iberia, AirEuropa, Abanca, and Banc Sabadell as customers, and NeuralTrust's own funding announcement corroborates the same four, and the company reports doubling its full-year 2025 annual recurring revenue in Q1 2026, though it disclosed no dollar figures. A strategic partnership with KPMG, plus collaborators including Capgemini and Sopra Steria, gives it a channel into large enterprise accounts.

The capital now matches the enterprise ambition more closely than before. NeuralTrust raised a $20 million seed led by Alstin Capital, with VentureFriends, Seaya, Kibo Ventures, and Banc Sabadell participating, a sizable round for a European cybersecurity startup. The plan is to deepen product integration, grow engineering, and expand across Europe while growing its international presence.

NeuralTrust does not publish list prices, but states enterprise pricing is based on the number of protected apps and agents, traffic volume, and deployment model, per its homepage FAQ. The rate card and contract mechanics remain undisclosed. A vendor that names the charged dimensions without a public rate card usually targets large negotiated enterprise deals, which fits a customer base of mostly billion-dollar enterprises.

The open-core model implies a land-and-expand meter rather than a single list price. NeuralTrust offers TrustGate as an open-source download a team can run locally, then frames the upgrade around governance, compliance, and depth of control across many teams, which reads as a scale-based enterprise tier layered over a free entry point. The paid platform charges on the number of protected apps and agents, traffic volume, and deployment model, while only the exact rate card and contract terms remain undisclosed.

The inferable belief is that buyers pay for centralized control over agent traffic rather than for any single feature. Confirming the unit and whether consumption is capped would require a sales conversation, which the hidden-price posture and demo-request flow signal is the intended route.

NeuralTrust delivers its platform in deployment modes chosen to match compliance needs. It offers a SaaS option, a hybrid cloud model that keeps a SaaS control plane while the data plane stays privately hosted so data never leaves the customer boundary, and an on-premises or air-gapped mode for defence, government, and financial-sector buyers. That range lets a bank or agency run inspection of sensitive prompts inside its own perimeter.

The architecture centers the gateway as the enforcement point. TrustGate brokers every model, tool, and protocol call, TrustGuard ingests agent traffic from gateways, SDKs, browsers, sidecars, and log streams and applies allow, block, transform, or alert decisions depending on the integration, and TrustLens tracks agent posture. NeuralTrust states that it inspects a large daily volume of agent interactions and intervenes on the small share it flags as malicious, a self-reported operating figure rather than an independently audited one.

The delivery is product rather than managed service in fetched sources. No analyst-staffed operations layer or accountability-for-outcomes offering appears, so the customer configures and runs the controls itself. The deployment pages do publish operational performance figures, including sub-100 ms p95 added latency and 20k+ requests per second per node. Published uptime and support SLAs do not surface in fetched pages.

NeuralTrust holds its own information-security certification, a meaningful starting artifact for a product that inspects privileged AI traffic. The company states it achieved ISO/IEC 27001:2022 certification, the recognized information-security management standard a procurement team can ask to see. For a vendor whose gateway and runtime engine sit in the path of a customer's sensitive prompts, that attestation addresses an early security-review question.

Deployment choice carries much of the trust case. Because the hybrid mode keeps the data plane inside the customer's boundary and the open-source gateway can run in the customer's own cloud, a regulated buyer can avoid sending proprietary prompts to a vendor cloud at all. The air-gapped mode extends that to defence and government buyers who cannot use a SaaS control plane.

The attestation is enterprise-grade but table-stakes rather than a moat. ISO 27001 eases procurement without blocking a substitute, the cited record identifies no regulation mandating this product class, and a buyer whose traffic the platform inspects should still resolve data-handling, model-provider, and retention terms in a formal review beyond the certification.

NeuralTrust positions itself as the control layer that all enterprise agent traffic flows through rather than a point tool. By brokering every model, tool, and protocol call at the gateway and enforcing at runtime across every surface agents run on, it aims to be, in its own words, the layer enterprise AI runs on, a claim that rests on becoming the single chokepoint for agent activity.

The open-source core is the adoption strategy and the structural weakness at once. Offering the open-source core is meant to win installs and a place in customer traffic, and the gateway brokers calls to any model so a customer is not locked to one provider, but an open gateway is also the layer a platform vendor could study and bundle into a deal an enterprise already signs. The durable separation is occupying the enforcement point first, not the code.

The ecosystem play leans on systems integrators rather than a marketplace. NeuralTrust names KPMG as a strategic partner and Capgemini and Sopra Steria as collaborators, relationships that can route it into large enterprise accounts, while the platform does not own the underlying models or tools it routes and governs.

NeuralTrust pairs commercial and operational pedigree with a research bench, and the reviewed sources document no prior security-sector exit, though the funding release notes COO Alejandro Domingo previously scaled a startup to an exit. CEO Joan Vendrell led AI deployment and security as Chief Data Officer at Mango after roles at Amazon and McKinsey, a leadership signal that is commercial and operational more than a track record of shipping security products. The co-founders join him from consulting and regulated-industry security roles, with CTO Victor Garcia having led security architecture in regulated telecom and banking and COO Alejandro Domingo having led AI transformation as an Associate Partner at McKinsey.

The research output is the team's strongest public asset. NeuralTrust documented the Echo Chamber multi-turn jailbreak and the Semantic Chaining multimodal attack, both reported as incorporated into the OWASP AI Security Project taxonomy, a sustained and externally checkable stream of in-domain adversarial work for a company this young.

The backer bench reinforces the signal. The $20 million seed drew Alstin Capital as lead, with VentureFriends, Seaya, Kibo Ventures, and Banc Sabadell participating. That volume of venture backing in a single seed round is itself a signal about how investors read the team.