18 open-source AI Security vendors.

Open-source software for AI Security that you can self-host, with auditable source code — regardless of the developer's headquarters.

• Arize AI / PhoenixNon-EU vendor

  Arize AI (Berkeley, US) builds the Arize AX platform and the open-source Phoenix library for LLM observability and evaluation, with an EU data-residency option.

  United States

• DeepchecksNon-EU vendor

  Open-source ML validation platform from Israeli company Deepchecks (Tel Aviv) providing comprehensive data and model testing from research to production.

  Israel

• Evidently AINon-EU vendor

  Open-source ML and LLM observability framework for evaluating, testing, and monitoring AI systems and data pipelines with 100+ built-in metrics; backed by a real company, not a controlling-company-free project.

  United States

• Garak (NVIDIA)Non-EU vendor

  Open-source LLM vulnerability scanner from NVIDIA's AI Red Team probing for hallucinations, data leakage, prompt injection, and jailbreaks across 50+ probe modules. Controlled by NVIDIA, not a community-governed project.

  United States

• GiskardEU sovereign

  French open-source platform for AI testing, red-teaming and guardrails.

  France

• Guardrails AINon-EU vendor

  Open-source Python framework for validating and structuring LLM outputs with 50+ pre-built validators and an optional managed service with observability dashboards.

  United States

• Langkit (WhyLabs)Self-hosted / open source

  LangKit is WhyLabs' open-source toolkit for monitoring LLMs, extracting signals such as toxicity, PII, sentiment and jailbreak detection from prompts and responses; WhyLabs discontinued its commercial platform but LangKit remains available as open source.

  United States · owner in United States

• Llama Guard (Meta)Non-EU vendor

  LLM-based safety classifier from Meta with customisable content moderation for text and image understanding across modalities. Controlled by Meta (Llama Community License), not a community-governed project.

  United States

• LLM Guard (Protect AI)Self-hosted / open source

  LLM Guard is an open-source security toolkit from Protect AI (now part of Palo Alto Networks) offering input/output scanners against prompt injection, data leakage and harmful language in LLM applications.

  United States · owner in United States

• Mistral ShieldstralSelf-hosted / open source

  3B-parameter policy-adaptive safety classifier released by Mistral AI (France) under Apache 2.0 as open weights, for on-device content moderation. Mistral AI is a real company, so not a controlling-company-free project despite the open licence.

  France

• NVIDIA NeMo GuardrailsNon-EU vendor

  Open-source toolkit from NVIDIA for adding programmable guardrails to LLM conversational systems, with support for self-hosted deployments. Controlled by NVIDIA, not a community-governed project.

  United States

• Portkey AISelf-hosted / open source

  Portkey AI (San Francisco, US; acquired by Palo Alto Networks in 2026 for Prisma AIRS) is an open-source AI gateway with 60+ guardrails against jailbreaks, prompt injection and PII leakage.

  United States · owner in United States

• Presidio (Microsoft)Self-hosted / open source

  Presidio is Microsoft's open-source framework for detecting, redacting and anonymizing personal data (PII) in text, images and structured data, widely used to prevent PII leakage in LLM pipelines.

  United States · owner in United States

• Project AsagoSelf-hosted / open source

  Red Hat-initiated open-source project (August 2026, Apache 2.0) translating NIST AI RMF, OWASP LLM Top 10 and EU AI Act requirements into technical governance controls for AI agents; participants include Red Hat, NVIDIA, IBM Research, MIT Lincoln Laboratory, Microsoft and the Alan Turing Institute — no single controlling party.

  United States

• PromptfooNon-EU vendor

  Open-source MIT-licensed LLM evaluation and red-teaming framework for testing prompts, models, and RAG pipelines, backed by a real VC-funded company.

  United States

• PyRIT (Microsoft)Non-EU vendor

  Open-source Python Risk Identification Tool from Microsoft enabling red team exercises to proactively identify risks in generative AI systems with multi-modal attack support. Controlled by Microsoft, not a community-governed project.

  United States

• TaoQ AISelf-hosted / open source

  Applied AI research and engineering company specialising in AI security, agent red-teaming, and EU AI Act conformity with open-source security testing frameworks for intelligent systems.

  Netherlands

• VigilSelf-hosted / open source

  Vigil is an open-source Python library and REST API (by researcher deadbits/Adam Swanda) that scans LLM prompts and responses for prompt injection, jailbreaks and prompt leakage; experimental alpha status with limited active maintenance.

  United States