sonnylabs-sdk

SonnyLabs is a security and oversight layer for AI agents, chatbots, and assistants used in business settings. It is described as providing guardrails for AI and as applying zero trust to AI, with the stated goal of keeping clean requests flowing to those systems while stopping manipulation and risky behavior.

Its core functions are organized around four stages. Before launch, SonnyLabs runs red teaming against an AI system and simulates attacks such as manipulation, data extraction, role hijacking, and agent abuse. In production, it records every conversation, request, and action an AI tries to take, presenting the activity in a searchable and exportable live dashboard. It also blocks manipulation attempts, hidden instructions, dangerous tool calls, risky actions, and confidential data or personally identifiable information leaving the model. For oversight and compliance, it can generate evidence packs, export audit logs, and provide ready mappings for EU AI Act article 15, SOC 2 control mapping, and vendor questionnaires based on the NIST AI RMF.

The product page names CISOs, security leaders, CTOs, CIOs, AI leaders, and compliance and governance teams among its intended audience, and it also groups use cases by industry and by role. It says the service plugs into existing security tools and can be used for business operations, conversational AI chatbots, sales AI agents, marketing automation, customer support, and investment AI agents. A quickstart, docs, playground, and interactive demos are listed for developers.

SonnyLabs is offered in cloud, self-hosted, and airgapped forms. It is presented as research-backed at University College Dublin and as ready for the EU AI Act.