Lightweight open classifier models that detect prompt injection and jailbreak attempts in LLM inputs.
Llama Prompt Guard 2 is Meta's family of small classifier models for detecting prompt injection and jailbreak attempts before text reaches a large language model. It comes in 86-million and 22-million parameter versions built on mDeBERTa encoders.
The models screen inputs of up to 512 tokens with low latency, making them suited to high-throughput guardrails. They are released under the Llama 4 Community License as part of Meta's Llama security tooling.
| Attribute | Llama Prompt Guard 2 | Security-One | |
|---|---|---|---|
| Provider |  |  | |
| Founded | 2004 | 2024 | |
| Sells To | Consumers, Enterprise | Enterprise, Small Business | |
| Pricing Model | Advertising | Recurring, SaaS, Usage-based | |
| Ownership | Public | Venture-backed |
Provider
Founded 2004
Sells To Consumers, Enterprise
Pricing Model Advertising
Ownership Public
Provider
Founded 2024
Sells To Enterprise, Small Business
Pricing Model Recurring, SaaS, Usage-based
Ownership Venture-backed
Security platform for protecting generative AI and LLM applications in enterprises.
Open-source toolkit for adding programmable guardrails to large language model applications.
AI/LLM Security Assessment tests LLM applications and infrastructure for prompt injection, data poisoning, and model extraction risks.
Amazon Bedrock Guardrails is a configurable safeguard layer that evaluates and filters inputs and outputs of generative AI applications.
Open-weight 8-billion-parameter model that verifies factual claims and flags hallucinations in AI-generated text.
AI/ML Security Services assess AI systems for adversarial attacks, model theft, prompt injection, and data integrity risks.
AI/ML Penetration Testing red-teams machine learning systems using the MITRE ATLAS and OWASP LLM frameworks.
An open-source AI gateway and proxy that calls many large language model APIs in the OpenAI format.
Platform securing machine learning models and AI systems against attacks.
NeuroWave is an enterprise platform that audits and corrects large language model behavior in real time.
Open LLM Leaderboard ranks open language models on reproducible public evaluation tasks.
Lakera Guard is a runtime security product for large language model applications.
Security platform for protecting generative AI and LLM applications in enterprises.
Open-source toolkit for adding programmable guardrails to large language model applications.
AI/LLM Security Assessment tests LLM applications and infrastructure for prompt injection, data poisoning, and model extraction risks.
Amazon Bedrock Guardrails is a configurable safeguard layer that evaluates and filters inputs and outputs of generative AI applications.
Open-weight 8-billion-parameter model that verifies factual claims and flags hallucinations in AI-generated text.
AI/ML Security Services assess AI systems for adversarial attacks, model theft, prompt injection, and data integrity risks.
AI/ML Penetration Testing red-teams machine learning systems using the MITRE ATLAS and OWASP LLM frameworks.
An open-source AI gateway and proxy that calls many large language model APIs in the OpenAI format.
Platform securing machine learning models and AI systems against attacks.
NeuroWave is an enterprise platform that audits and corrects large language model behavior in real time.
Arx is Fabraix's run-time defence layer that blocks prompt injection and memory poisoning in AI agents.
Open LLM Leaderboard ranks open language models on reproducible public evaluation tasks.
Veritas is a family of open-source language models fine-tuned to reduce factual hallucination.
