Abnormal AI vs Alice

About Alice

Alice's WonderSuite gives security, product, and compliance teams red teaming and real-time guardrails. ActiveFence Ltd does business as Alice. WonderSuite is the AI security platform for teams launching customer-facing AI in regulated industries. WonderSuite tests security and safety posture before launch, protects the AI experience in production, and catches what changes over time. WonderSuite combines automated adversarial testing with expert-led red teaming and real-time protection. WonderSuite covers pre-launch red teaming, real-time guardrails, and continuous post-launch testing in a single platform. In WonderSuite, automated red teaming assesses how an agent behaves before any guardrails are deployed. WonderSuite aligns every test, guardrail, and finding to customer policies and to compliance frameworks, including OWASP, NIST, MITRE ATLAS, and the EU AI Act.^[2][3][4][5]

WonderBuild red teams customer-facing AI apps, agents, and agentic workflows before launch. WonderBuild runs thousands of adversarial tests based on custom policies that surface prompt injections, jailbreaks, data leakage risks, and compliance and regulatory gaps. WonderBuild uses adversarial data from Rabbit Hole to expose the path attackers take. WonderBuild identifies gaps between policies and the system prompt, across text, multimodal, and agentic systems. WonderBuild gives CISO, legal, and compliance teams test documentation with findings mapped to regulatory requirements. WonderBuild red-teaming findings integrate into WonderFence.^[6]

WonderFence sits between external-facing AI and users, intercepting harmful, non-compliant, and off-policy responses before they land. WonderFence deploys runtime guardrails tuned to the customer's application and policies, intercepting harmful, non-compliant, or off-brand responses. WonderFence reviews every prompt coming in and every response going out, blocks harmful inputs before they hit the LLM, and blocks harmful outputs in real time before they impact users. WonderFence trains a dedicated detector for each policy using real-world adversarial data, deployed at sub-100ms latency. WonderFence watches every message in both directions, 24/7, blocking, masking, or alerting in real time, with activity tracked in a dashboard. WonderFence maps guardrails to the regulatory frameworks that legal and compliance teams care about: EU AI Act, ISO 42001, NIST, MITRE ATLAS, and OWASP. WonderFence is now a Databricks Unity AI Gateway External Policy that can be attached once in Unity Catalog to screen every governed model call, with no SDK and no code changes.^[3][4][7][8]

WonderCheck runs continuous automated red teaming in production. WonderCheck's ongoing automated red-teaming detects drift, regression, and emerging risks. WonderCheck watches for drift, surfacing behavioral changes before they become incidents. WonderCheck integrates directly into CI/CD pipelines for on-demand, scheduled, and ongoing testing. WonderCheck ranks issues by severity and pairs each one with clear remediation guidance. WonderCheck findings flow directly into WonderFence guardrails, or back into WonderBuild for deeper refinement.^[4][7][9]

Rabbit Hole is an adversarial network of in-house researchers and data that connects online attacks with the real world. Rabbit Hole collects real attacks and uses them to secure AI. Alice's solutions are powered by Rabbit Hole, an adversarial threat intelligence engine built on billions of real-world data samples collected across nearly a decade. Alice produces safety, security, multimodal, agentic, and skills-based datasets for model, application, and agent training. Alice combines expert-led and automated red-teaming to stress-test models across text, image, audio, and video under real adversarial conditions. Alice's evaluation and benchmark datasets are based on Alice's harms taxonomy or customized to customer policies. Alice provides AIGC and harmful content learning sets purpose-built for guardrail and content classifier creation. Alice provides simulated, high-fidelity environments and scenarios for training and testing AI agents and agentic models. Alice and AI21 post-trained NVIDIA's open Nemotron 3.5 Super to resist indirect prompt injection. Alice and Inflection AI are releasing a multi-turn red-teaming dataset covering risks such as emotional dependency and misplaced belief.^{[8][10][11][12]}

Alice names foundation model labs, global enterprises and platforms deploying AI as its customers. WonderSuite provides end-to-end security coverage for agentic systems, from adversarial testing before deployment to real-time behavioral guardrails in production. WonderSuite covers prompt injection, unsafe tool calls, data leakage, and multi-step exploits across the entire agentic architecture. Alice connects directly through existing third-party Customer Experience (CX) agent extensions, so there is no need to swap vendors or rebuild integrations. Alice goes past literal translation to catch regional slang, dialect, euphemism, and coded phrasing. Alice watches every message in both directions, blocking, masking, or alerting in real time, at P99 under 120ms.^{[2][3][10][12]}

Alice's ActiveFence is a UGC trust and safety layer that detects, prevents, and responds to online harm. ActiveFence live feeds surface harmful content, viral narratives, and coordinated manipulation in real time, across languages and regions. ActiveFence delivers datasets of adversarial indicators, domains, IPs, and hashes. ActiveFence uncovers vulnerabilities across code, systems, and infrastructure. ActiveFence conducts deep web investigations to expose networks, specific threat actors, and emerging tactics. ActiveFence's SaaS tools are built for content moderation. ActiveMind, powered by in-house large language models, helps moderation teams validate, flag, and respond to harmful content.^[8][13]