Alice Raises $140M to Secure AI Models and Agents

Alice spent most of its life studying the parts of the internet companies wish did not exist. Fraud networks, extremist communities, coordinated manipulation, and online abuse became the raw material for understanding how attackers behave before the rest of the market learned to call the problem AI security.

That history now has $140M of new financing behind it. Apax Digital Funds led Alice's August 25, 2026, round, with Samsung, SentinelOne, Maj Invest, MoreTech, and Phoenix Insurance joining alongside existing investors Resolute Ventures, Grove Ventures, CRV, Highland Europe, Vintage Investments, Norwest, NFX, and Claltech. The round brings Alice's total funding to $280M, and Apax Digital will join the board.

The capital lands as AI moves from generating answers to taking actions. A model connected to email, files, databases, and software tools does not merely create a content-moderation problem when it fails. It can create an operating event, which is why the market around testing, governing, and monitoring AI behavior is beginning to look like infrastructure rather than an optional safety feature.

What Alice Raised

Alice did not disclose a formal round label or valuation. CTech reported a valuation between $700M and $800M, but that figure should be treated as third-party reporting rather than company disclosure. The official announcement identifies Apax Digital as the lead, lists the new and returning investor group, and says the new financing takes total capital raised to $280M.

The company previously operated as ActiveFence. In 2021, ActiveFence announced $100M across a Series B and a previously undisclosed Series A. CRV and Highland Europe led the Series B, while Grove Ventures and Norwest led the Series A. Reporting around the 2026 rebrand from ActiveFence to Alice put prior funding at $140M, although the reviewed sources do not provide a complete round-by-round explanation for the additional $40M. The safest accounting is the official one: $140M in the current round and $280M in total funding.

From Trust and Safety to AI Security

Alice was founded in 2018 by Noam Schwartz, Iftach Orr, Alon Porat, and Eyal Dykan. Schwartz remains CEO, Orr is CTO, and Porat is Chief Customer Officer. Dykan is a verified co-founder but is not listed on Alice's current leadership page, so no current operating role should be inferred.

The company first built its business around online trust and safety, tracking malicious behavior across the open web before it reached large consumer platforms. Years of that work became Rabbit Hole, which Alice describes as the world's largest proprietary dataset of real-world adversarial and harmful content. The dataset gives Alice a library of behavior drawn from actual fraud, manipulation, extremism, and abuse rather than relying only on synthetic test cases.

That lineage is the center of the investment thesis. AI attackers change tactics, models change behavior, and enterprise policies differ by customer. A reusable list of prohibited words cannot govern a system that retrieves private information, uses tools, calls APIs, and makes decisions across a workflow. Alice is betting that a continuously refreshed map of adversarial behavior can produce better testing and defense as those systems evolve.

How WonderSuite Works

WonderSuite covers 3 stages of the AI lifecycle. WonderBuild stress-tests models, applications, and agents before launch for problems such as jailbreaks, prompt injection, data leakage, and policy violations. WonderFence evaluates prompts and responses in production against organization-specific policies. WonderCheck continues adversarial testing after launch to find drift, regressions, and vulnerabilities introduced by model, prompt, or application changes.

Rabbit Hole connects those products. Attacks observed in the wild can become new test cases. Test results can inform runtime controls. Production behavior can expose the next gap that needs to enter the research loop. The business value depends on whether that loop improves faster than attackers can change their methods, not simply on whether Alice possesses a large archive.

Alice's current documentation presents the platform as model-agnostic and built for customer-facing GenAI applications, agents, and foundation-model workflows. That matters because a generic model safeguard cannot know every bank's data policy, every healthcare company's escalation rule, or every software agent's permitted actions. The enforcement layer has to understand the application, the user's context, and what the system is authorized to do.

The Operating Signals Behind the Round

Alice says it is approaching $100M in annual recurring revenue and that its AI business grew more than 500% over the last two years. The company also reports more than 150 AI-security researchers, protection covering more than 3B people online, and work with 8 of the 10 leading AI model labs. Named model-lab relationships include Anthropic, Google, and Cohere, while the company's historical platform work includes Google, Meta, TikTok, and Amazon.

Those metrics come from the company and investor announcement, not an independent audit. They still explain why Apax is treating Alice as a scaled growth investment rather than an early research bet. Alice has a large existing trust-and-safety operation, a growing AI business, and a customer base that gives it exposure to both pre-deployment model testing and production risk.

The investor mix reinforces that positioning. A security company such as SentinelOne and a technology manufacturer such as Samsung have strategic reasons to care about the controls surrounding models and agents. Apax Digital brings growth-equity experience and will join the board, placing the investor closer to the execution required to convert research depth into repeatable enterprise revenue.

What the $140M Is Supposed to Change

Alice plans to deepen the technology that tests, defends, and monitors AI systems across their lifecycle. The company will also expand the team behind Rabbit Hole so the dataset can keep up with changing attack techniques and scale its go-to-market organization for foundation-model labs and enterprises building with AI.

The product obligation is demanding. Pre-launch testing has to find failures worth fixing without burying teams in noise. Runtime guardrails have to enforce policy without breaking the user experience. Ongoing testing has to catch meaningful drift as models, prompts, retrieval sources, and permissions change. A defense that works only in a controlled evaluation is not enough when the model enters a live workflow.

The commercial obligation is just as sharp. Enterprises already buy application security, identity, data protection, governance, and monitoring tools. Alice has to show where AI-specific controls fit without becoming another dashboard that detects problems but cannot change outcomes. The company must translate adversarial intelligence into evidence that security, legal, product, and engineering teams can use to approve, operate, and improve AI systems.

What This Funding Signals for Agentic AI

The financing is a market signal about permission. Chatbots made unsafe output visible. Agents make unsafe behavior consequential because they can touch the systems where companies keep money, customer data, communications, and operational authority. The security question changes from “what might the model say?” to “what can this system see, decide, call, store, and do?”

Alice enters that market with an unusual asset: nearly a decade spent watching how people exploit communicative technology at scale. The $140M round does not prove that Rabbit Hole will become the default security layer for agentic AI, and the company's operating metrics remain self-reported. It does give Alice more resources to test the proposition in production, where the value of adversarial intelligence is measured by the failures customers avoid and the authority they can safely give software.

The worst behavior on the internet is not disappearing. It is being translated into prompts, tool calls, synthetic identities, and automated workflows. Alice's opportunity is to make that history useful before an agent turns somebody else's old trick into a new operating incident.