Alice WonderSuite AI Red Teaming and Guardrails

Alice is the brand adopted in January 2026 by ActiveFence Ltd, a trust and safety company founded in 2018 and led by co-founder and CEO Noam Schwartz. With about 350 employees, most of them in Israel, and 140 million USD raised from investors including Grove Ventures, CRV, Highland Europe and Norwest, it shifted from protecting users of social platforms toward securing foundation models, AI applications and agents after large customers asked for that capability.

WonderSuite covers three stages. WonderBuild performs automated and managed pre-launch red teaming against jailbreaks, prompt injection and data extraction and produces evidence for launch approval. WonderFence applies real-time guardrails to prompts and responses tuned to each application's policies. WonderCheck runs continuous adversarial tests in production to catch regressions after model or prompt changes. The Rabbit Hole engine feeds adversarial intelligence into classifiers covering over 100 languages and multiple modalities. Logos include Amazon, Google, Meta, TikTok, Nvidia and Cohere.

• WonderBuild: automated and managed pre-launch red teaming for jailbreaks, prompt injection and data extraction.
• WonderFence: real-time guardrails on prompts and responses with custom policies.
• WonderCheck: continuous adversarial testing in production to catch regressions.
• Rabbit Hole: adversarial intelligence engine feeding detection classifiers.
• Multilingual and multimodal: coverage in over 100 languages, model-agnostic.
• Compliance alignment: resources mapped to the EU AI Act, NIST and OWASP.

Vietnamese e-commerce and social apps can red team AI shopping assistants and filter harmful content in Vietnamese and other regional languages. Banks and insurers can test customer-facing chatbots before launch and keep guardrails in production. Edtech platforms can apply child-safety policies to AI tutors.

Product, security and legal teams get documented red teaming evidence before launching customer-facing AI. Guardrails tuned to each application's own policies reduce harmful outputs without, according to Alice, meaningful latency impact. Coverage of over 100 languages suits multilingual markets where many tools only handle English. Continuous production testing catches regressions after every model or prompt update.

Consumer-facing AI assistants, companions and agents are spreading quickly, and harmful or manipulated outputs create legal and reputational risk. Trust and safety expertise built for social platforms is being applied to GenAI, and Southeast Asian platforms serving many languages need safety tools that work beyond English.

ActiveFence was founded in 2018 and rebranded as Alice in January 2026; the legal entity remains ActiveFence Ltd and activefence.com redirects to alice.io. It has raised 140 million USD and says it protects over 3 billion users on social platforms. The site shows SOC 2, HIPAA, ISO 27001 and ISO 42001 badges.