Alice

• No benchmark, audit, or third-party evaluation of Rabbit Hole’s coverage against prompt injection and agent misuse appears in the public record, so a buyer cannot verify detection quality before purchase.
• A platform vendor such as Microsoft, AWS, or NVIDIA could bundle comparable runtime guardrails into its AI stack, compressing the standalone market Alice sells into.
• Most traction evidence is vendor-published and groups clients with partners, so no cited source separates paying customers from partnerships.
• The cited record names no funding round since the 2021 Series B, so a repositioning into a new category is being financed without a disclosed raise.
• Alice publishes different runtime latency figures on different surfaces, below 100 milliseconds on its WonderFence page and below 150 milliseconds on its AWS Marketplace listing, so a buyer sizing capacity has no single vendor number to plan against.

Alice sells AI security to the companies that build and deploy generative AI, a buyer facing prompt injection, jailbreaks, and harmful model output at production scale. Its homepage frames the exposure as application-level exploitation reaching applications, agents, and integrations.

Independent coverage confirms the repositioning rather than the pain. Calcalist reported the January 2026 rebrand and growing demand for AI safeguards, while the EU AI Act, ISO 42001, NIST, MITRE ATLAS, and OWASP references come from Alice’s own pages and are its positioning rather than measured buyer demand.

The framing shift is recent. Alice spent most of its history protecting social platforms from abuse and harmful content, and it now presents the same adversarial data as the basis for AI security. Its product pages make that link explicit, so a buyer evaluating the pivot is testing execution rather than intent. [s1, s6, s3, s9]

WonderSuite is Alice’s AI security platform, organized as three lifecycle products. WonderBuild red teams models, apps, agents, and agentic workflows before launch with thousands of adversarial tests written against the customer’s own policies. WonderFence applies runtime guardrails that evaluate every prompt and every response. WonderCheck keeps testing production systems to catch drift and emerging risks.

Rabbit Hole is the engine underneath the suite. Alice describes it as adversarial intelligence gathered over a decade, spanning billions of samples and more than 120 languages, and its WonderBuild page ties that data to attack simulation against a customer’s own models and agents.

The documented depth is concrete and vendor-attested. A three-stage lifecycle, per-policy detectors trained on adversarial data, multimodal coverage in more than 20 languages, and a stated sub-100-millisecond runtime latency all come from Alice’s own pages, with no third-party technical evaluation in the cited record. [s2, s3, s9, s4]

Alice entered AI guardrails and red teaming carrying an eight-year content-safety business rather than starting from the AI problem. That history is what its pages lead with.

The differentiator Alice presses is Rabbit Hole. It says it built the corpus by collecting adversarial and abusive interactions for a decade across more than 120 languages, and it leans on that depth when it claims accuracy and coverage.

What the public record does not settle is how the corpus performs on model-security threats. Alice’s own pages describe it as covering prompt injection and jailbreaks alongside toxic content, and no benchmark, audit, or third-party evaluation of that coverage appears in the cited record, so a buyer cannot check the claim without running its own test. [s4, s9, s2]

Alice’s named-relationship evidence is specific. The rebrand release lists Amazon, TikTok, NVIDIA, Cohere, and Black Forest Labs together as clients and partners, without saying which are paying customers, and Alice states that it protects seven of the ten biggest AI foundation models.

Testimonials attach people to the claim. Cohere’s head of AI safety and a responsible-AI lead at Amazon AGI are quoted on Alice’s site describing its evaluation and safety work, which a reader can check against those named individuals rather than against an unattributed logo wall.

A cloud channel now carries part of the motion. Alice lists Real-Time Guardrails on AWS Marketplace with contract-based pricing and a 4.9 rating from 14 reviews. The rest of the traction evidence is vendor-published, so a buyer weighing it has 14 outside data points and a set of vendor references. [s5, s6, s1, s10]

Alice was founded in 2018, and its about page lists Noam Schwartz as co-founder and chief executive, Iftach Orr as co-founder and chief technology officer, and Alon Porat as co-founder and chief customer officer.

Credibility comes from sustained delivery to demanding buyers. Alice’s about page says it works with top tech giants, global social platforms, and seven of the ten leading AI foundational models, and it kept those relationships through the AI rebrand, which is harder to fake than a single launch.

What the cited record does not show is a prior founder exit or a disclosed funding round since the 2021 Series B. The team’s depth is in trust-and-safety operations, and its standing in model security is still being established. [s7, s6, s8]

Alice presents enterprise trust signals appropriate to its buyers. SOC 2 and ISO 27001 certification badges appear on its homepage, and WonderFence maps a customer’s guardrails to the EU AI Act, ISO 42001, NIST, MITRE ATLAS, and OWASP.

The compliance posture matches the buyer. Selling runtime guardrails to regulated enterprises and frontier labs requires exactly these attestations, and Alice’s eight years serving large platforms make the posture plausible.

The signals come from the vendor’s own pages rather than an independently published audit report or a customer-accessible trust portal, so nothing in the cited record lets an outside reader inspect the underlying certifications. [s1, s3]

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

Alice sells to enterprises and AI labs putting customer-facing generative AI into production, and it aims squarely at regulated industries. The buyer is the security or product team that owns an AI experience, and Alice frames the work as testing that experience before launch, protecting it in production, and catching what changes over time.

The segment skews to large, exposed buyers. The rebrand announcement names Amazon, TikTok, NVIDIA, Cohere, and Black Forest Labs as clients and partners and states that Alice protects seven of the ten biggest AI foundation models, organizations for whom a single unsafe model output carries real brand and regulatory exposure.

Alice now sorts its pitch by industry as well as by use case. Its site routes buyers through child-facing products, financial services, healthcare, and insurance, so a sales conversation can start from the buyer's vertical instead of a general AI-safety argument.

Alice’s product is WonderSuite, a lifecycle platform organized as three stages. WonderBuild red teams apps, agents, and agentic workflows before launch using thousands of adversarial tests written against the customer’s own policies. WonderFence evaluates every prompt and every response in production. WonderCheck keeps testing after launch to catch drift and regressions as models change.

The runtime detail is specific. WonderFence trains a dedicated detector for each customer policy on real-world adversarial data and covers text, image, audio, and video in more than 20 languages. Alice publishes a latency figure below 100 milliseconds on the WonderFence page and below 150 milliseconds on its AWS Marketplace listing, so a buyer planning capacity gets a different number depending on which page it reads.

Rabbit Hole is the asset underneath the suite. Alice describes it as adversarial intelligence collected over a decade, spanning billions of samples and more than 120 languages, and its product pages tie that data directly to testing for prompt injection, jailbreaks, and data leakage. Every efficacy claim attached to it comes from Alice’s own pages.

The cited record shows Alice reaching buyers through longstanding relationships, rebrand-driven press, and an AWS channel, and it does not document Alice’s outbound motion. Recasting an eight-year trust-and-safety business as an AI-security company drew independent press coverage that reintroduced it to a different buyer.

Its traction evidence is named and partly press-carried. The rebrand release lists Amazon, TikTok, NVIDIA, Cohere, and Black Forest Labs as clients and partners and states that Alice protects seven of the ten biggest AI foundation models. Cohere’s head of AI safety and a responsible-AI lead at Amazon AGI are quoted on Alice’s site, which attaches named people at named organizations to the claim.

A second route now runs through a cloud channel. Alice lists Real-Time Guardrails on AWS Marketplace as a contract-priced service, where it carries a 4.9 rating from 14 reviews on a review surface Alice does not host. Fourteen reviews is a thin sample, and no other buyer feedback in the cited record comes from a surface Alice does not curate.

The reviewed Alice product pages publish no rate card. The platform page directs prospects to a booked demo or a downloadable overview, with no list price, per-unit figure, or free tier, which is consistent with a negotiated enterprise sale to large security and product organizations.

Its AWS Marketplace listing is the one place a public dollar figure appears. The listing shows a 12-month annual commitment of 500,000 dollars against a description telling buyers to contact the company for actual pricing, and it states that the offer is intended to be private and that self-service purchases are refunded automatically, so the number opens a negotiation rather than closing one.

That gives an outside reader an order of magnitude and not a comparison. A prospect weighing Alice against an AI-safety capability bundled into a platform it already licenses still cannot compare a negotiated quote to the marginal cost of that bundled feature.

Alice delivers software its customer’s own teams operate, with expert help attached. WonderSuite is a platform a security or product team runs to test, protect, and monitor its AI, and Alice’s platform page describes the combination as automated adversarial testing plus expert-led red teaming and real-time protection.

WonderFence sits in the live flow of prompts and responses. It inspects every prompt going in and every response coming out, which raises the availability and speed bar Alice has to clear before a frontier lab routes production traffic through it.

The recurring work is monitoring rather than a one-time engagement. WonderCheck keeps re-testing production systems as models change, and WonderBuild returns findings ranked by severity with remediation guidance, so the work the product hands back is a queue the customer’s own engineers then work through.

Alice displays the enterprise attestations its buyers expect. SOC 2 and ISO 27001 certification badges appear on its homepage, the baseline assurance a frontier lab or regulated enterprise asks for before routing sensitive prompts and outputs through an outside service.

Its products are positioned around regulatory alignment. The WonderFence page says the product maps a customer’s guardrails to the EU AI Act, ISO 42001, NIST, MITRE ATLAS, and OWASP, and that every decision is logged and every policy documented, which is the audit trail its regulated buyers are themselves measured against.

The signals are appropriate to the stage and self-displayed. The attestations and framework mappings appear on Alice’s own pages rather than in an independently published audit report or a customer-accessible trust portal, so they read as credible vendor claims that an external report would strengthen.

Alice positions WonderSuite as a lifecycle platform rather than a single tool, unifying pre-launch testing, runtime protection, and production monitoring under one suite fed by a shared intelligence engine. A buyer can standardize several AI-safety controls on one vendor instead of stitching point products together.

Its ecosystem strength runs through data rather than a developer marketplace. Alice splits its offerings into a Wonder family for AI systems and an Active family for user-generated content, and both draw on the same adversarial corpus, so the connective tissue across the portfolio is that shared data.

The outward connections are a cloud channel and a research arm rather than third-party builders. Alice sells Real-Time Guardrails through AWS Marketplace and promotes Alice Labs as frontier research for AI labs. No partner-built integration marketplace appears on the pages fetched here, so the platform claim rests on the breadth of the suite and the data beneath it.

Alice was founded in 2018 and is led by co-founder and CEO Noam Schwartz, with co-founder Iftach Orr as chief technology officer and co-founder Alon Porat as chief customer officer. Three founders still hold executive roles, which is a sustained operating record rather than a team assembled for the AI wave.

The credibility comes from years of delivery to demanding buyers. Alice’s about page says the company works with top tech giants, global social platforms, and seven of the ten leading AI foundational models, and its CEO told Calcalist that Amazon, TikTok, and governments asked for the technology to protect their users.

The funding is established but not recent. Alice has raised about 140 million dollars since 2018 from investors including CRV, Highland Europe, Norwest Venture Partners, and Grove Ventures, and the most recent round the cited record names is the Series B reported in 2021. That is a long stretch without a disclosed raise for a company repositioning into a new category.