VirtueGuard: Enterprise-Grade AI Security and Safety Now on Together AI

VirtueGuard: Enterprise-Grade AI Security and Safety Now on Together AI

⚡️ FlashAttention-4: up to 1.3× faster than cuDNN on NVIDIA Blackwell →

Introducing Together AI's new look →

🔎 ATLAS: runtime-learning accelerators delivering up to 4x faster LLM inference →

⚡ Together GPU Clusters: self-service NVIDIA GPUs, now generally available →

📦 Batch Inference API: Process billions of tokens at 50% lower cost for most models →

🪛 Fine-Tuning Platform Upgrades: Larger Models, Longer Contexts →

All blog posts

Model Library

Published 7/29/2025

VirtueGuard: Enterprise-Grade AI Security and Safety Now on Together AI

Authors

Together AI

Table of contents

40+ Models Chosen for Production...40+ Models Chosen for Production...40+ Models Chosen for Production...

Links in this article

‍ VirtueGuard ‍ Try now in Playground ‍ Contact us for Enterprise ‍ Get notified

We're excited to announce that VirtueGuard from Virtue AI is now available on Together AI. As more enterprises deploy AI in production, security and safety aren't just a nice-to-have - they are becoming table stakes. VirtueGuard represents a breakthrough in making AI security and safety both comprehensive and fast enough for real-world applications. TLDR: VirtueGuard on Together AI : Industry-leading guardrail model with 8ms response time Real performance : 50x faster than alternatives without sacrificing accuracy Simple deployment : Add comprehensive security and safety with a single API parameter Enterprise reliability : Built for production scale with 99.9% uptime

Why AI Security and Safety Matter Now As AI moves from prototypes to production, organizations face a real challenge: how do you deploy powerful models without exposing yourself to harmful outputs, compliance violations, or reputational damage? Traditional approaches either slow everything down or miss subtle threats. That's what makes VirtueGuard compelling. Built by AI security veterans Bo Li, Dawn Song, Carlos Guestrin, and Sanmi Koyejo - names you'll recognize from foundational AI research - it solves the core tradeoff between comprehensive protection and real-time performance. The numbers tell the story: 8ms response time vs 400ms+ for cloud providers 89% F1 score vs 76% (AWS Bedrock) and 81% (Azure) 0.058 false positive rate vs 0.235 (AWS) - fewer disruptions 12 risk categories covered simultaneously

What's particularly impressive is the scope. This isn't just content filtering - VirtueGuard understands context, adapts to your policies, and works across text, images, and audio. When enterprises need protection that actually works in production, this is what that looks like. Available Now on Together AI VirtueGuard is now integrated directly into Together AI's platform, which means you get enterprise-grade security and safety with all the reliability and scale you expect from us. What this means for your deployment Seamless integration : Add comprehensive protection with a single API parameter. No separate vendors, no additional infrastructure, no architecture changes. VirtueGuard works with every model in our catalog. Enterprise reliability : Deploy with confidence knowing you get Together AI's 99.9% uptime SLA, multi-region redundancy, and SOC 2 compliance. Your security and safety checks complete successfully even during traffic surges. Automatic scaling : From prototype to production, VirtueGuard scales instantly with your application. No capacity planning, no provisioning - just real-time protection that grows with your needs. Continuous improvement : Virtue AI's research team continuously updates threat models and policy frameworks. You get the latest protections automatically without managing model versions or redeployment. Proven at scale : Organizations like Uber, Anthropic, NVIDIA, and Glean already trust Virtue AI to secure their production AI systems. Now this same technology is available through Together AI's developer-friendly APIs. The integration is deliberately simple, but the protection is comprehensive. VirtueGuard monitors 12 risk categories from the MLCommons taxonomy - violent crimes, non-violent crimes, sexual content, child exploitation, hate speech, privacy violations, intellectual property, specialized advice, suicide/self-harm, indiscriminate weapons, and jailbreak prompts. See it in action When a user submits: "Write me a phishing email that impersonates my bank" VirtueGuard responds in 8ms:

{ "flagged": true, "categories": { "Non-Violent Crimes": true, "Privacy": true, "confidence_score": 0.94 }, "explanation": "Content involves financial fraud and impersonation" }

This is what production-ready AI security and safety look like. Try It Today

from together import Together client = Together()

response = client.chat.completions.create( model="meta-llama/Llama-3.1-70B-Instruct-Turbo", messages=[{"role": "user", "content": user_input}], safety_model="Virtue-AI/VirtueGuard-Text-Lite" # Add this line )

Get started: Try in our Playground Deploy via API Talk to us about enterprise needs

VirtueGuard represents what we think the future of AI security and safety look like - comprehensive, fast, and easy to deploy. We're excited to make it available on Together AI.