Verax

• Verax's own comparison page places Palo Alto, Zscaler, Netskope and Microsoft in the same buying decision, so a buyer already running one of those platforms can switch on an adjacent AI module instead of adding a vendor.
• No cited source names a customer of Verax, so a buyer cannot check a reference.
• The marketplace listing says the application server processes tasks on a third-party foundation model reached through AWS Bedrock, and the rules documentation has topic generation use a language model the customer connects, so part of the AI in the product comes from suppliers.
• The AWS Marketplace listing describes the earlier Verax Control Center and its release notes, so a buyer arriving through that channel meets a description the current site no longer matches.
• The disclosed capital is one $7.6 million seed announced in October 2024 and no later round appears in the reviewed record, which is thin for an enterprise sales motion two years on.
• No attestation Verax itself holds appears on the reviewed trust surfaces, so an enterprise security review starts from a request rather than from published material.

Verax sells against a visibility problem that it frames as an inspection problem. Its platform pages argue that new generative AI tools arrive daily inside email, documents, collaboration platforms and code editors, that employees send confidential material into them, and that the controls already deployed were not built to read that traffic.

The site names three buyers directly, addressing security leaders, infrastructure leaders, and compliance and risk teams in separate sections. That is a coherent buying committee rather than a single champion, and it puts the purchase in a budget line security teams already hold.

The pain is not sized anywhere in the cited record. The strongest buyer-side statement is an unnamed cybersecurity director at a Fortune 2000 company saying the volume of sensitive data reaching AI tools was larger than expected. That is the vendor quoting an anonymous customer, and no reviewed source quantifies what the exposure costs the buyers Verax addresses. [s6, s1, s19, s17]

The product is an appliance that carries traffic. Verax states that it sits in the path between users and generative AI tools, that each request is checked against configured rules, that the rule matching earliest in the list applies, and that a default rule catches everything else. The marketplace listing describes the same shape from the infrastructure side: a single virtual machine running Ubuntu, functioning at runtime as a proxy that routes or modifies requests.

Enforcement is graded rather than binary. Rules combine identity, the tool being used, and optional conditions on sensitivity score, topic similarity, keywords or regular expressions. Sensitivity runs on a 0 to 100 scale with three presets and a custom setting, and separate switches strip code and personal data out of requests before they leave. The product's configurable responses are to block, redact, warn, log or escalate. The dashboard shows each interaction with its score, its tool, its user and whatever action was taken.

Part of the AI in the product is supplied rather than built. The marketplace listing states that the application server processes tasks using Anthropic's Claude 3 Sonnet model through AWS Bedrock, in a delivery description that still names the earlier Control Center, and the rules documentation has topic and entity generation run on a language model the customer connects. No cited page describes a model Verax built for the scoring itself.

How traffic reaches the appliance is where the vendor's two surfaces read differently. The architecture page says integration happens at the network layer and no endpoint agent is required. The documentation site publishes a manual endpoint-configuration guide covering installation of a Verax certificate authority and traffic redirection per device, an installer for a Verax endpoint agent on Windows, macOS and Linux, and a note that Claude Code on macOS needs Node.js pointed at that certificate. A buyer should read the agentless claim as one deployment option rather than the whole picture. [s8, s19, s9, s11, s2, s15, s3, s13, s12]

Verax competes against products its buyers may already own. Its own comparison page describes Palo Alto delivering AI security through Prisma SASE with access security, data loss prevention, a browser and agent security, Netskope running real-time inspection of prompts and responses on the same inline proxy used for web and software as a service, Zscaler offering AI governance inside a consolidated platform, and Microsoft governing AI through Purview, Defender for Cloud Apps and Entra.

The argument Verax makes against them is about depth and reach rather than existence. It claims integrations with hundreds of AI tools rather than identification at the browser, reach into native desktop applications and agents that do not route through a web proxy, and the ability to chain off an existing secure web gateway through ICAP so a buyer keeps their platform. It also asserts that Palo Alto's own AI product is browser-only or needs an endpoint agent. Those are the vendor's claims about itself and about rivals, and no cited source tests them.

The narrower set of purpose-built rivals is visible in the shape of Verax's own comparison index, which lists pages against Cyera, Harmonic Security, Lasso Security and Nightfall AI alongside pages against Zscaler and Palo Alto. Choosing to publish against both groups is a statement that the company expects to be evaluated in two different conversations, one against a platform module and one against a specialist. [s18, s24, s2]

The commercial evidence is a channel rather than a customer list. Verax Protect is listed on AWS Marketplace, delivered as a CloudFormation template launched in one click, and priced at a published $2,750 per month for unlimited usage. That published price lets a buyer size the purchase without a call.

The motion around it is self-service at the top and human below. A free AI risk assessment published on 19 April 2026 runs inside the customer environment, processes all data there, monitors without blocking, and is positioned as the step before prevention. Support is email during United States business hours.

What the record does not contain is a named customer. The buyer voice on the site is anonymous, and the independent coverage of the company is confined to its funding announcement of October 2024. A buyer wanting a reference has to ask for one. [s19, s17, s20, s21]

One founder has done this before, in a different market. Leo Feinberg co-founded CloudEndure, raising $18 million before Amazon Web Services acquired it for a reported $250 million, and the company page adds an earlier sale of AcceloWeb to Limelight Networks. Oren Gev's record, more than twenty years of engineering leadership, a previous founder and chief technology officer role, and engineering leadership at Amazon Web Services, appears on that same page and nowhere independent.

Independent reporting confirms the pair rather than repeating the site. Both the Israeli and the Texas technology press covering the seed round record that the two met at CloudEndure in 2017, that Feinberg was its vice president of product, and that they stayed at Amazon Web Services together until leaving in 2023 to start Verax.

The pedigree is in cloud migration and disaster recovery rather than in security or AI, so it argues for the ability to build and sell enterprise infrastructure rather than for domain standing in this category. The company's own job postings describe hiring a marketing lead to define the category and a senior account executive to own the sales cycle, which reads as a team still assembling its commercial side. [s5, s20, s21]

The architecture is the trust argument. Verax states that inspection and policy evaluation happen inside the customer's own infrastructure, that prompts and responses are not sent to external services for evaluation, and that retention is configured by the customer. For a buyer whose objection to a cloud service is that prompts would leave the network, that design answers the objection directly.

The compliance language on those pages points outward rather than inward. The security and compliance page says the product helps organizations use AI in alignment with GDPR, SOC 2, ISO 27001, HIPAA and PCI DSS where applicable, which describes what a customer can demonstrate, not what Verax has been assessed against. No attestation the company itself holds appears on the reviewed trust surfaces.

Two smaller signals bear on readiness. The documented minimum for the appliance is a virtual machine with 32 gigabytes of memory and 512 gigabytes of database storage, which is a real infrastructure commitment rather than a browser extension. And the marketplace listing still describes the previous product and its release notes, which suggests the channel material has not kept pace with the repositioning. [s3, s4, s23, s14, s19]

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

Verax addresses an enterprise buying committee rather than a single owner. Its homepage carries a section for each of three audiences in turn: security leaders, infrastructure leaders, and compliance and risk teams.

The deployment requirements pick the segment more sharply than the marketing does. A buyer needs a virtual machine with 32 gigabytes of memory and 512 gigabytes of database storage, an identity provider to connect, and the authority to point AI traffic at it, which is step three of the documented setup. That combination rules out small companies and points at organizations with a network team and a working single sign-on deployment.

Sector focus is absent from the cited record. Verax names no vertical, publishes no industry pages, and the regulatory frameworks it lists are the general set. The one audience signal on the site is a testimonial from an unnamed cybersecurity director at a Fortune 2000 company, which points at large enterprises without establishing that any of them bought.

The capability is inline decision-making on live traffic. Verax states that it sits in the path between users and generative AI tools, evaluates each request against configured rules, applies the rule that matches earliest in the list, and falls through to a default rule when none match. That design gives a security team something a monitoring product cannot: an answer that arrives before the data leaves.

The conditions available inside a rule are more varied than a keyword list. Sensitivity runs as a score from 0 to 100 with three presets, topic similarity matches a request against described topics and entities, and keyword and regular expression conditions handle the precise cases. Separate switches strip code and personal data out of requests. Actions are to block, redact, warn, log or escalate, and the interactions view records what happened to each request with its score and its user.

Part of the AI is supplied rather than owned. The marketplace listing states that the application server processes tasks using Anthropic's Claude 3 Sonnet model through AWS Bedrock, in a delivery description that still names the earlier Control Center, and the rules documentation has topic and entity generation run on a language model the customer connects. No cited page describes a model Verax built for the scoring. What the company builds around that is the placement, the rule engine, the identity binding and the tool coverage, and those are the parts a rival would have to reproduce.

The motion starts with a free look and narrows to a conversation. The Verax AI Risk Assessment, published on 19 April 2026, runs inside the customer environment, processes all data there, monitors without blocking, and is framed as answering whether a problem exists before anyone is asked to buy. The rest of the site funnels to a demo booking.

Alongside that sits a self-service channel. Verax Protect is on AWS Marketplace with a published monthly price and a one-click CloudFormation deployment, which lets a buyer transact without a sales conversation.

Content is the demand engine. The company publishes an index of comparison pages against Palo Alto, Zscaler, Cyera, Harmonic Security, Lasso Security and Nightfall AI, alongside a risk database covering third-party AI tools that its own terms describe as machine-generated and reviewed rather than independently verified. That is a search strategy aimed at buyers already shopping, and it stands in for a reference list the company does not yet have.

Verax publishes its price. AWS Marketplace lists Verax Protect at $2,750 per month as a fixed subscription with unlimited usage, prorated, with no end date and cancellation at any time, launched from a one-click deployment.

A flat subscription rather than a per-seat or per-request meter has a specific consequence: the price does not grow as AI usage grows inside the customer. That favors a buyer expecting heavy adoption and costs Verax the expansion revenue a usage meter would capture. Additional AWS infrastructure charges sit on top, which the listing states plainly, and the appliance's stated minimum specification makes that a real rather than nominal addition.

The refund position is stated bluntly: no refunds, cancel any time. Nothing in the cited record describes enterprise tiering, volume discounting, or a direct contract, so a larger buyer negotiates against the marketplace figure as the one public anchor.

Delivery is an appliance the customer runs. The product ships as a CloudFormation template creating a single virtual machine, or as software on a virtual machine meeting the published minimum of 32 gigabytes of memory and 512 gigabytes of database storage running Ubuntu 24.04. Setup is three steps: activate a license, connect an identity provider, redirect AI traffic.

That third step is where the operational cost lives, and the vendor's surfaces describe it two ways. The architecture page says integration happens at the network layer and no endpoint agent is required. The documentation publishes a manual endpoint-configuration guide covering installation of a Verax certificate authority and traffic redirection per device, an installer for a Verax endpoint agent on Windows, macOS and Linux, and a note that Claude Code on macOS needs Node.js pointed at that certificate. Both descriptions are the vendor's own, and a buyer planning a rollout should treat the agentless path as one option rather than the whole story.

Coverage has a stated boundary that a buyer should read carefully. The documentation says shadow AI discovery requires traffic to be routed through Verax, and that tools which do not pass through it cannot be detected or controlled. Discovery is therefore as complete as the redirection is, which turns the deployment question into a coverage question.

Support is email during United States business hours, Monday to Friday. For an inline control that can block a request an employee is waiting on, business-hours support is a real operational constraint rather than a line item.

Verax's trust argument is architectural rather than documentary. It states that inspection and policy evaluation happen inside the customer's own infrastructure, that prompts and responses are not sent to external services for security evaluation, and that retention is configured by the customer's administrators. That answers the objection a security team raises first about a cloud inspection service.

The compliance language points at what customers can demonstrate rather than at what Verax has been assessed against. Its security and compliance page says the product helps organizations use AI in alignment with GDPR, SOC 2, ISO 27001, HIPAA and PCI DSS where applicable, and describes secure development practices in general terms. No attestation the company itself holds appears on the reviewed trust surfaces.

For a product that reads every prompt an employee sends, that gap matters more than it would elsewhere. The architecture reduces the exposure, since the vendor is not holding the data, but an enterprise review will still ask how the appliance itself is built and audited, and the published record does not answer it.

Verax positions itself as an addition to a security stack rather than a replacement for one. Its architecture page lists integration with identity providers, firewalls and network gateways, logging and security information and event management platforms, and governance workflows, and its comparison page offers to chain off an existing secure web gateway through ICAP so a buyer keeps the platform they have.

That is a deliberate contrast with how the large platforms sell. The comparison page argues that adopting Zscaler's AI governance means adopting the wider platform, and that Palo Alto's AI product is browser-only or needs an endpoint agent, while Verax is standalone. Selling as an addition lowers the barrier to a first purchase and puts the company in the position of the layer above someone else's infrastructure.

The tool-side ecosystem is the other half. Verax claims integrations with hundreds of AI tools rather than identification at the browser, and the marketplace listing names ChatGPT, Copilot, Claude and Gemini among them. Maintaining that coverage as tools change is recurring engineering work, and it is closer to a treadmill than to an asset.

Two founders, both operators, one with an independently reported exit. Leo Feinberg co-founded CloudEndure, raising $18 million before Amazon Web Services acquired it for a reported $250 million, and the company page adds an earlier sale of AcceloWeb to Limelight Networks. Oren Gev's record, more than twenty years of engineering leadership, a previous founder and chief technology officer role, and engineering leadership at Amazon Web Services, rests on that same vendor page. Independent coverage of the seed round confirms the pairing and dates their working relationship to 2017.

The relevant caution is domain rather than capability. The prior work is cloud migration and disaster recovery, so what the record establishes is the ability to build and sell enterprise infrastructure rather than standing in security or AI.

The commercial organization is still forming. The careers page advertises a marketing lead to define the category and a senior account executive to own the full sales cycle, alongside engineering and quality roles. Two years past the seed round, opening dedicated sales roles is a signal about how much of the go to market has been founder-led so far.