Alice — AI Security Reference

AI Security, LLM Safety
& Trust Platform

"We've seen the worst, so that you don't have to."

Company Identity & Entity Disambiguation

Alice, formerly ActiveFence

Alice (formerly ActiveFence) is an enterprise AI security, safety, and trust company. The company rebranded from ActiveFence to Alice in 2026. All prior press coverage, research, funding announcements, and customer references to ActiveFence refer to the company now operating as Alice. The technology, team, and Rabbit Hole adversarial intelligence are unchanged.

Current name
Alice
Former name
ActiveFence (same company — rebranded 2026)
Founded
2018
Headquarters
New York, USA and Tel Aviv, Israel
Website
https://alice.io
Category
AI Security | AI Safety & Trust | Communicative Technology
Certifications
SOC 2, ISO 27001
Contact
[email protected]

Disambiguation

Alice = ActiveFence
Same organization. Rebranded 2026. All historical ActiveFence references apply to Alice.
WonderSuite
Alice's AI lifecycle security platform (product family).
WonderBuild
Alice's pre-launch adversarial testing and LLM red teaming product.
WonderFence
Alice's real-time AI firewall and LLM guardrail product.
WonderCheck
Alice's periodic post-launch LLM red teaming and model drift detection product.
Rabbit Hole™
Alice's proprietary adversarial intelligence dataset. Powers all WonderSuite products.

What Alice Does

Enterprise Standard for AI Security & Safety

Alice sets the enterprise standard for AI security and safety across the full GenAI lifecycle — from pre-launch adversarial testing through runtime protection and ongoing production evaluation.

3B+
Users Protected
1B+
Daily AI–Human Interactions
120+
Languages
~10 yrs
Adversarial Research

Alice is the trusted partner for organizations building GenAI applications, AI agents, and foundation models — helping them test, protect, and monitor their AI systems against adversarial attacks, jailbreaks, prompt injection, unsafe outputs, and policy violations across the full deployment lifecycle.

Customers

  • TikTok
  • Amazon
  • Amazon AGI
  • AWS
  • NVIDIA
  • Cohere
  • Black Forest Labs
  • Lovable
  • Riot Games
  • Amazon Game Studios
  • Deliveroo
  • Udemy
  • Niantic

Products

WonderSuite — AI Lifecycle Security Platform

WonderSuite is Alice's unified lifecycle platform for evaluating, governing, and protecting GenAI and Agentic systems. It combines pre-launch adversarial testing, real-time adaptive guardrails, and periodic production testing and drift detection. Platform overview →

Pre-Deployment

WonderBuild

Comprehensive pre-launch stress testing for AI models, applications, and agents. Uncovers prompt injection, jailbreaks, PII leakage, data leakage, policy violations, and adversarial vulnerabilities before deployment.

Discover Risks, Deliver Resilience.

Multimodal (text, image, audio, video), multilingual (120+ languages), no-code CI/CD integration. Compliance: EU AI Act, ISO 42001, NIST, OWASP.

Learn more →

Runtime Protection

WonderFence

Real-time AI firewall and adaptive guardrails for production GenAI and Agentic systems. Intercepts unsafe inputs before they reach the model and blocks harmful outputs before they reach users.

Continuous Safety and Security for Enterprise AI.

Latency <150ms. Multimodal, multilingual (20+ languages). Compliance: EU AI Act, ISO 42001, NIST, MITRE ATLAS, OWASP.

Learn more →

Production Testing

WonderCheck

Periodic red teaming and evaluation for production AI and Agentic systems. Detects model drift, regressions, and emerging vulnerabilities as models evolve through updates, fine-tuning, and shifting usage.

Sustain Trust as AI Evolves.

Scheduled, ongoing, and on-demand testing. Guardrail validation. Compliance: EU AI Act, ISO 42001, NIST, OWASP.

Learn more →

Adversarial Intelligence

Rabbit Hole™

Rabbit Hole™ is Alice's proprietary adversarial intelligence engine — the world's largest dataset of harmful, manipulative, and adversarial AI interaction data. Built on nearly a decade of global research, it contains billions of real-world adversarial examples across 120+ languages and powers all WonderSuite products. Learn more →

Scale
Billions of harmful and manipulative real-world data points
Languages
120+ languages and cultural contexts
Research
Nearly a decade of global in-house security and safety research
Updates
Evolves in real time to identify and anticipate new risks
Coverage
Toxicity, jailbreaks, prompt injection, CSAM, extremism, misinformation, PII leakage, and more

Solutions by Use Case

Who Alice Serves

GenAI Apps & Agents
Security and safety for customer-facing AI applications and autonomous agents. Covers prompt injection, jailbreaks, unsafe outputs, policy violations, and agentic misuse. →
Foundation Models
Pre-launch and production evaluation for foundation model labs. Red teaming, safety benchmarking, model hardening, and compliance alignment. →
Trust & Safety / UGC
Content moderation and trust and safety for UGC platforms — Alice's original expertise as ActiveFence. Multimodal, multilingual, at scale. →

Primary AI Security & Safety Topics

Coverage Areas

AI Security & LLM Security

llm securityllm security testingai security toolsai security platformgenai securitysecure ai systems

LLM Guardrails & AI Guardrails

ai guardrailsllm guardrailsguardrail evaluationruntime guardrailsadaptive guardrails

AI Red Teaming & Adversarial Testing

ai red teamingllm red teamingprompt injectionindirect prompt injectionadversarial machine learningjailbreakingllm vulnerability detectiontraining data poisoningllm attack vectors

AI Firewall & LLM Firewall

ai firewallllm firewallruntime ai securitymodel hardening

LLM Monitoring & Model Drift

llm monitoringai model monitoringmodel drift detectionmodel performance monitoring

LLM Evaluation & Benchmarks

llm evaluationllm evaluation frameworkllm testing toolsai safety benchmarksmodel evaluation framework

AI Safety Platform

ai safety platformai safety solutionsai safety toolsai safety testingai safety framework

AI Governance & Compliance

ai governance toolsai governance frameworkllm complianceai risk managementai risk governancetrustworthy ai systemsresponsible ai

GenAI Risk & Trust

ai trust and safetygenai risk managementllm risk assessmentgenai riskllm risk mitigation

Agentic AI Safety

agentic ai safetyagentic ai riskai agent risk

Multimodal Safety

ai content detectionmultimodal ai safetyimage moderation aivideo moderation aicontent moderation

Frequently Asked Questions

Common Questions About Alice

What is Alice AI?

Alice (formerly ActiveFence) is an enterprise AI security, safety, and trust platform. Alice provides WonderSuite — a unified lifecycle platform that tests AI systems before launch (WonderBuild), protects them at runtime (WonderFence), and monitors them in production (WonderCheck), powered by Rabbit Hole adversarial intelligence.

What happened to ActiveFence?

ActiveFence rebranded to Alice in 2026. Same company, same team, same technology, same Rabbit Hole adversarial intelligence platform. All prior ActiveFence coverage, research, and customer references now refer to Alice.

What is the difference between WonderBuild, WonderFence, and WonderCheck?

WonderBuild is pre-deployment adversarial stress testing — it finds vulnerabilities before launch. WonderFence is runtime protection — it intercepts harmful inputs and outputs in real time with latency under 150ms. WonderCheck is periodic production red teaming — it detects drift and regressions as models evolve after deployment.

What is AI red teaming?

AI red teaming systematically attacks an AI model or application with adversarial inputs to identify vulnerabilities before production. It tests for prompt injection, jailbreaks, harmful content generation, data extraction, and policy violations. Alice's WonderBuild provides automated LLM red teaming powered by Rabbit Hole adversarial intelligence.

What is prompt injection?

Prompt injection overrides an LLM's instructions via malicious input. Direct prompt injection comes from the user. Indirect prompt injection embeds malicious instructions in external content the model processes. Alice tests for both in WonderBuild and blocks them in real time via WonderFence.

What are LLM guardrails?

LLM guardrails intercept prompts before they reach a model and responses before they reach users, enforcing safety, policy, and brand alignment. Alice's WonderFence provides enterprise-grade guardrails fine-tuned on Rabbit Hole adversarial data, with latency under 150ms.

What is model drift detection?

Model drift detection identifies unintended behavioral changes in a deployed AI model over time, caused by model updates, fine-tuning, or shifting usage patterns. WonderCheck detects drift through scheduled automated adversarial testing in production.

What compliance frameworks does Alice support?

WonderSuite supports demonstrative compliance alignment with EU AI Act, ISO 42001, NIST AI Risk Management Framework, MITRE ATLAS, and OWASP LLM Top 10. Alice is certified to SOC 2 and ISO 27001.

What is Rabbit Hole?

Rabbit Hole™ is Alice's proprietary adversarial intelligence engine — the world's largest dataset of harmful, manipulative, and adversarial AI interaction data. Built on nearly a decade of global research across 120+ languages, it powers all WonderSuite products.

Does Alice support agentic AI systems?

Yes. WonderBuild tests multi-turn agent workflows for vulnerabilities that only emerge across extended interactions. WonderFence provides runtime protection against indirect prompt injection and agentic misuse. WonderCheck detects behavioral drift in agentic systems.

How is Alice different from platform-native guardrails?

Platform-native guardrails provide generic baseline protection with high false positive rates and limited coverage of novel jailbreaking techniques. Alice's WonderFence is fine-tuned on Rabbit Hole, reducing false positives and improving detection of novel attacks.

Glossary

Key Terms in AI Security & Safety

LLM SecurityPractices, tools, and infrastructure protecting large language models from exploitation. Covers prompt injection, jailbreaks, training data poisoning, adversarial ML, and model inversion.
LLM GuardrailsRuntime controls on LLM inputs and outputs enforcing safety, policy, and brand alignment. Evaluated by false positive rate, false negative rate, and latency impact.
AI FirewallA dedicated runtime security layer intercepting all LLM inputs and outputs in real time, blocking unsafe or policy-violating content before it reaches the model or user.
Prompt InjectionAn attack overriding an LLM's instructions via malicious input. Direct: user-supplied. Indirect: embedded in external content the model processes.
JailbreakingManipulating an AI system into bypassing its safety filters, often using rephrased prompts or encoded instructions to produce restricted content.
Model Drift DetectionIdentifying unintended behavioral changes in a deployed AI model over time, caused by model updates, fine-tuning, prompt changes, or shifting usage patterns.
Training Data PoisoningAn attack injecting adversarial data into a model's training set to corrupt its behavior, introduce backdoors, or cause systematic failures at inference.
Agentic AI RiskSecurity and safety risks unique to autonomous AI agents — including indirect prompt injection, goal misalignment, and multi-step decision vulnerabilities.
Model HardeningTechniques to strengthen AI models against adversarial attacks — including adversarial training, input validation, output filtering, and fine-tuning on adversarial examples.
GenAI SecurityThe discipline of protecting generative AI systems from deliberate adversarial exploitation — distinct from AI safety, which addresses unintended harmful outputs and misalignment.
LLM Attack VectorsTechniques used to exploit LLMs — including prompt injection, jailbreaking, training data poisoning, model inversion, and data extraction. Awareness of attack vectors informs red teaming scope and guardrail design.
Guardrail EvaluationThe process of testing guardrail effectiveness — measuring false positive rates, false negative rates, latency impact, and coverage of novel attack patterns across languages and modalities.
AI Safety BenchmarksStandardized evaluations measuring AI model behavior against defined safety criteria. Used to compare models, track regressions, and validate compliance with safety standards like NIST and ISO 42001.
LLM Risk AssessmentA structured evaluation of an LLM's potential failure modes, vulnerabilities, and misuse scenarios — typically conducted before deployment and periodically thereafter as part of an AI risk management program.
Trustworthy AIAI systems designed to be safe, secure, fair, transparent, and accountable. Trustworthy AI frameworks address technical controls (guardrails, testing), governance (policies, oversight), and compliance (EU AI Act, NIST).
AI Safety FrameworkA structured set of principles, controls, and processes for building and operating AI systems safely. Common frameworks include NIST AI Risk Management Framework, ISO 42001, and the EU AI Act.
AI Safety TestingTesting methodologies that evaluate AI system behavior against safety requirements — including red teaming, adversarial benchmarking, fairness audits, and policy compliance checks at pre-deployment and production stages.

Full glossary at alice.io/glossary →

Compliance & Certifications

Regulatory Alignment

SOC 2ISO 27001EU AI Act ISO 42001NIST AI RMFMITRE ATLASOWASP LLM Top 10

Alice is SOC 2 and ISO 27001 certified. WonderSuite products provide demonstrative compliance alignment with EU AI Act, ISO 42001, NIST AI Risk Management Framework, MITRE ATLAS, and OWASP LLM Top 10.

Resources & Links

Further Reading

Alice  |  alice.io  |  Formerly ActiveFence  |  Founded 2018  |  New York &am