Introducing DynamoGuard On-Device Guardrails: The Most Secure Generative AI Solution

Enterprise AI adoption is often stifled by data governance risks: what happens when sensitive data proliferates throughout AI-enabled Copilots and embedded AI features. Users can expose LLMs to sensitive data stored on their local endpoint devices, where traditional data loss protection (DLP) controls struggle to govern data transfer into AI Copilots. Moreover, as enterprises begin equipping their workforces with AI PCs, individual employees are now empowered with their own local LLMs, but lack the traditional oversight and auditability of today’s cloud LLMs. This leaves several major security and compliance gaps across the enterprise. 

DynamoGuard On-Device Guardrails deliver the most secure and compliant GenAI solution, empowering teams to harness the full potential of GenAI, while enforcing real-time compliance checks and data loss controls.  Powered by Dynamo’s proprietary lightweight models, DynamoGuard On Device enables secure, real-time guardrailing directly on-device. 

DynamoGuard On-Device

DynamoGuard On Device sets a new standard for GenAI security by proving:

1. Real-Time On Device Guardrailing: Guardrails run directly on local NPU, CPU, or GPU resources, meaning that sensitive user inputs never leave the device. By running guardrails directly on your devices, DynamoGuard eliminates the risk of data exposure and ensures compliance with even the most stringent security standards. 
2. Unparalleled Performance: Dynamo’s ultralightweight guardrails enable low latency guardrailing and minimal power consumption. With guardrailing latencies of less than 55ms and power consumption of under 1% for up to 1k guardrail requests, your teams can stay productive without compromise.
3. Flexible Deployment: DynamoGuard On Device can integrate with enterprise applications in two deployment modes to enable guardrailing of both On Device and Cloud AI solutions. DynamoGuard offers seamless integration, regardless of whether you’re enhancing the security of cloud-based tools or implementing a fully local GenAI solution.

How it Works: The Technology Powering DynamoGuard On Device

DynamoGuard On Device is powered by a set of key developments that ensure unmatched efficiency and performance: 

1. Ultralightweight, Optimized Guardrail Models: DynamoGuard leverages Small Language Models (SLMs) that are specifically designed to run efficiently. These models are further optimized to run On Device through techniques like quantization, enabling the model to run seamlessly on GPU, CPU, or NPU resources while preserving performance. 
2. Efficient, Unified Inference Engine: PRISMA is Dynamo’s proprietary inference engine that powers DynamoGuard. It’s responsible for managing inference requests and efficiently allocating resources to meet latency requirements while delivering high performance.It also offers direct integration with leading on-device compute frameworks across mobile and PC platforms, optimized for efficient AI execution.
3. Run 100+ Guardrails On-Device: DynamoGuard leverages a proprietary guardrail multiplexing algorithm to efficiently run over 100 unique policy guardrails on-device — all without sacrificing latency. Powered by a purpose-trained model and PRISMA, it significantly reduces storage and memory overhead while delivering fast, scalable policy enforcement across both mobile and PC-class platforms.

Deployment Scenarios

Hybrid Deployment

In the hybrid deployment scenario, DynamoGuard is used to secure sensitive inputs on-device before any data is transmitted to cloud-based tools like Microsoft 365 Copilot. 

Imagine an employee working with a cloud-based AI tool like Microsoft 365 Copilot. Without proper security measures, sensitive data could inadvertently be sent in a prompt to the application. With DynamoGuard On Device, inputs to the AI tool are guardrailed locally, ensuring that no confidential information is transmitted while maintaining seamless integration with the cloud tool.

Fully On-Device Deployment

In the fully on device deployment scenario, DynamoGuard is integrated with local GenAI applications also running on-device. DynamoGuard ensures that all interactions remain secure by enforcing guardrails directly on the device, eliminating the need for any cloud dependency while upholding enterprise compliance policies.

Why Choose DynamoGuard On-Device?

• Enterprise-Grade Security: DynamoGuard On-Device is the ultimate GenAI security solution for enterprises, enabling compliance with stringent security and compliance regulations
• Efficient Guardrailing: DynamoGuard delivers robust AI guardrailing with minimal overhead, ensuring fast performance and low energy consumption.
• Customizable Integration: Whether securing cloud tools or enhancing local applications, DynamoGuard integrates seamlessly into your GenAI tools.

Get Started Today

As GenAI adoption accelerates, the need for secure, efficient solutions will only grow. DynamoGuard On Device provides the most secure Generative AI solution for your enterprise – enabling it to innovate securely. 

Ready to see how your enterprise can leverage DynamoGuard On Device? Request a demo today to see DynamoGuard On Device in action or explore how our on-device optimizations can seamlessly integrate into your cloud solutions.