Safe Labs AI

Safe Labs AI is an open-source red-teaming framework for AI agents. It is built to help find vulnerabilities in agent systems before they are attacked, and it is described as aligned with the OWASP Agentic Top 10. The tool generates audit-ready compliance reports and supports LangChain, CrewAI, AutoGPT, and custom AI agents.

Its core workflow is centered on 47 adversarial test cases mapped to the OWASP Agentic Security Initiative, spanning ASI01 through ASI10. It also says the framework systematically exercises an agent, logs each interaction, and grades responses automatically. Users can choose pre-built OWASP ASI suites or compose custom attack sequences, and the quickstart shows a Python API and a command-line interface for running scans against an agent endpoint.

Reporting is another explicit part of the product. Safe Labs AI can export results as PDF, JSON, or SARIF, and the page describes the output as including severity rankings, reproduction steps, and remediation guidance. The examples also show a structured findings file with scan metadata, severity counts, and reproduced findings. Use cases named on the page include AI startups, security teams, red teamers, and enterprises in regulated industries such as banking, healthcare, and government. Those audiences are linked to pre-launch safety checks, compliance audits, and integration into existing toolchains.

The tool is delivered as a package installable with pip using safelabs-eval, and the quickstart shows a default safelabs command for running suites. The page identifies version v0.1.1, the Apache 2.0 license, and GitHub availability. It also says Safe Labs AI is built by Safe Labs AI Inc. and offers early access through a waitlist.