Alice, a company focused on trust, safety, and security for artificial intelligence, announced that it has raised $140 million in a new funding round. The goal is straightforward: help secure AI systems against fast-evolving weaknesses and adversarial behavior, including prompt manipulation and jailbreak-style attacks.
This latest investment brings Alice’s total funding to $280 million. According to the announcement, the company plans to use the capital to advance its AI platform, grow the team behind its proprietary intelligence dataset, and scale its go-to-market efforts with partners across the AI ecosystem.
Why Alice is investing in stronger AI defenses
Security for AI models is not a one-time task. As models move from development into real-world use, new attack paths appear—often shaped by how users interact with systems. Alice’s approach is built around finding those failure modes early and then applying protection continuously as models operate at scale.
The company works with AI developers to stress-test foundational models before public release. This includes research designed to surface unusual or unstable behaviors that can be triggered through adversarial inputs or complex agent-like workflows.
Pre-release stress testing and adversarial probing
In Alice’s research division, the team uses mock malicious inputs and tests models with challenging tasks that resemble real attempts at misuse. The intent is to identify vulnerabilities such as prompt injection and jailbreak attempts before systems are deployed broadly.
By repeatedly probing how models respond under hostile conditions, Alice aims to harden model behavior and reduce the risk of erratic outputs. This is particularly important for organizations deploying foundational models, where small instruction changes can sometimes lead to large shifts in behavior.
Runtime guardrails for production AI
Pre-release testing is only one part of the security picture. Once models enter production, Alice provides a layer of ongoing protection through continuous red-teaming and dynamic runtime guardrails.
These guardrails are designed to help align system behavior with an organization’s operating requirements. Rather than relying on a single fixed policy, the platform supports ways to adapt rules as needs change—especially when different business units require different constraints.
Organizations can also configure internal guidelines, simulate potential breaches, and monitor traffic in real time. That combination helps teams detect suspicious behavior patterns while they observe how the system actually performs under day-to-day conditions.
Rabbit Hole: using threat intelligence to intercept hostile actions
A key piece of Alice’s platform is a proprietary database called Rabbit Hole. The company describes the repository as an archive built over nearly a decade by tracking digital fraud, ideological extremism, and manipulation campaigns.
Because the database contains records of harmful content and hostile actions, Alice’s platform can use that history to recognize and disrupt attacks targeting generative AI systems. In practical terms, it supports detection and interception workflows designed to reduce the chance that harmful requests lead to harmful outputs.
A dedicated research facility with 150+ specialists
Beyond product features, Alice maintains a dedicated research facility for studying vulnerabilities across machine learning architectures. The lab employs more than 150 specialists, and the team collaborates with major model developers to anticipate how systems can fail.
That work includes analyzing potential manipulation vectors and exploring weaknesses that could be exploited during real deployments. The aim is to close security gaps before attackers find reliable ways to leverage them.
Led by Apax Digital Funds as Alice expands
The funding round was led by Apax Digital Funds. Other participants include MoreTech, Phoenix Financial, Resolute Ventures, Grove Ventures, CRV, Highland Europe, Vintage Investments, Norwest, NFX, and Claltech.
Alice said the investment will also support the expansion of its proprietary intelligence dataset and the scaling of its market approach—suggesting the company plans to bring its security tooling and services to a broader set of enterprise and developer customers.
Democratizing security capabilities, not just model access
Alice’s CEO and co-founder, Noam Schwartz, emphasized a growing gap between how quickly people gain access to powerful AI capabilities and how rapidly defenses keep up. He noted that attackers can try countless variations, and that meaningful protection requires learning from what is actually being used to abuse technology.
According to Schwartz, the company has spent nearly a decade studying how people exploit digital platforms at global scale—and now applying those learnings to the ways adversaries probe AI models. In his view, Alice raises $140M as part of an effort to accelerate the availability of defenses ahead of the broader democratization of AI capabilities.
From ActiveFence to Alice
Before operating under its current name, Alice went by ActiveFence. The company’s headquarters are in New York and Tel Aviv.
With this round, the company signals continued momentum in the AI trust, safety, and security space—particularly around defenses that combine pre-release evaluation, production monitoring, and threat-informed detection.
What this means for organizations adopting AI
If you are deploying AI systems—especially generative models—security needs to extend beyond model selection. It should include testing before release, controls during runtime, and visibility into system traffic when the model is used in real workflows.
Alice’s focus on stress-testing, runtime guardrails, and intelligence-driven detection reflects that reality. With $140 million backing the effort, the company appears positioned to expand both its research capacity and the practical tooling enterprises can use to reduce risk.
Conclusion
Alice’s announcement that it raises $140M marks a significant push to strengthen AI model defenses. The company’s plan blends pre-release stress testing, continuous red-teaming and dynamic runtime guardrails, and threat intelligence enabled by its Rabbit Hole database. For organizations moving AI models into production, that combination aims to improve reliability against adversarial prompts and manipulation attempts as systems evolve.
