AI applications
AI Firewall
Once AI is in use, security has to keep up. One unified line of defense for your data center, AI Compute Appliances and AI applications.
What to defend against
External attacks
Scans, intrusions and abnormal traffic hit your public-facing services directly.
Prompt injection
Crafted inputs bypass the rules set for AI applications.
Sensitive data leaks
AI applications expose internal information in their answers that should stay private.
Abuse and unauthorized access
Unauthorized accounts or APIs consume model resources at scale.
Key capabilities
Access control
Set access policies by user, application and API.
Threat detection
Detect and block abnormal traffic and attacks.
LLM application protection
Detect prompt injection and filter sensitive content in inputs and outputs.
Logging and auditing
Complete records of access and calls to meet audit requirements.
Actual capabilities and specifications are subject to the final product proposal.
Deployment
Inline deployment
Sits at the service entry point for real-time detection and blocking.
Out-of-band deployment
Analyzes mirrored traffic without affecting existing links.
Integrated with the appliance
Delivered with the AI Compute Appliance, with protection out of the box.
Deployment and compliance
| Item | Details |
|---|---|
| Protected assets | Public-facing services, AI Compute Appliances, AI applications and model APIs |
| Deployment | Inline or out-of-band; can be delivered integrated with the AI Compute Appliance |
| Policy management | Policies by user, application and API, with unified logging and auditing |
| Sizing | Throughput and concurrency sized to your workload; see the solution proposal |
- Compliance documents and MLPS (China's Multi-Level Protection Scheme) alignment notes are provided at the proposal stage
- Deployment and audit recommendations based on your industry's regulatory requirements
FAQ
How is it different from a traditional firewall?
A traditional firewall mainly handles network-layer traffic and ports. The AI Firewall adds an understanding of LLM application inputs and outputs, detecting prompt injection, sensitive content leaks and API abuse. It is built specifically to protect AI applications and compute.
Can I buy it separately?
Yes. It can be delivered with the AI Compute Appliance, or deployed on its own in front of your existing AI applications or data center environment.
Does it meet MLPS requirements?
The AI Firewall can serve as part of the access control, intrusion prevention and security audit measures required under MLPS (China's Multi-Level Protection Scheme). How it fits your case is described in the proposal, based on your protection level and assessment requirements.
Related
Request a trial of AI Firewall
A consultant will reply during business hours (China Standard Time) with site, configuration and pricing recommendations.
- leo@hs.com.cn
- 24/7 technical support