0sec labs releases 0, an open-source AI security harness featuring web ui and cli

Zurich-based 0sec Labs has launched 0, a new open-source AI security harness designed to conduct vulnerability investigations, code reviews, pentesting, reverse engineering, and fix generation. This tool functions as a control plane around Large Language Models (LLMs) and security utilities, allowing users to run scoped workflows against authorized targets by plugging in a provider. The harness is available under the MIT or Apache-2.0 license as a research preview. 0 supports multiple access methods, including a local browser console, a headless Command Line Interface (CLI), and an MCP server. Model access is user-provided, supporting providers such as OpenAI, Anthropic, DeepSeek, Azure, and Grok, among others. For compiled targets, the system utilizes 0verse, an engine wrapping Ghidra, angr, and AFL++. The research team noted that the harness is self-evolving and multi-model. The project, backed by Y Combinator, has documented over 30 merged Linux kernel 0-days across various subsystems. While the core harness is free, a hosted version, 0cloud, is available via a waitlist, offering trial credits and per-test pricing. Users can explore the GitHub repository for details on this release.