AI Governance System (AGS) is an innovative solution that provides governance and management of LLM usage across the organizational workforce available via SaaS or deployable to your environment.Â
With AGS, your organization can harness the immense power of LLMs while ensuring the highest level of privacy and security. ZeroTrusted.ai’s robust LLM Firewall fortifies our solution, providing an extra layer of protection.Â
AGS offers a hassle-free solution that can be deployed on-premises or via Azure or AWS cloud-containerized virtual machines, allowing for swift deployment within minutes.Â
Identify and Stop:
Know what your organization is sending and receiving from LLMs/AI
Build keyword lists to block or anonymize
Control what your users do and view
FEATURES | Description | ZeroTrusted.ai's AGS | OpenAI Teams |
Team Users | |||
Multi Users | Multiple users can be added to one plan | Unlimted | 13 users |
Privacy | |||
Compliance Violation Detection & Prevention | Ensures all compliance violations are detection and prevention | ||
Anonymize Prompts | Maintain prompt anonymity when utilizing LLMs to ensure privacy. | ||
Keyword SafeGuard | Blocking custom words to prevent jailbreaks | ||
Streaming Data and Synchronous | Option to stream data for latency or synchronously to mitigate streaming data attack | ||
Disable History | Feature allowing user to disable history being saved | ||
No Data Harvesting | Prompt data can be used for training LLM | ||
Governance | |||
Compliance Violation Detection & Prevention | Monitor all compliance violations are detection and prevention across user base or enterprise wide | ||
Anonymize Prompts | Configure and enforce prompt anonymity when utilizing LLMs to ensure privacy across user base or enterprise wide | ||
Keyword SafeGuard | Configure and enforce blocking custom words to prevent jailbreaks across user base or enterprise wide | ||
Streaming Data and Synchronous | Option to stream data for latency or synchronously to mitigate streaming data attack | ||
Disable History | Enable or disable history being saved across user base or enterprise wide | ||
No Data Harvesting | Prompt data can never be used for training LLM | ||
LLMs | |||
Number of LLM Providers | Supports multiple 3rd party LLMs such as Claude, Cohere, PaLM, GPT-3.5, GPT-4, as well as custom LLMs such as the Llama series | 7 LLM providers | 1 LLM Provider |
Decentralization & Distribution | Use of multiple LLMs provide the option to choose LLM independent entities, rather than being controlled by a single organization | 5 LLMs (Parallel usage) | 1 LLM (Parallel usage) |
LLM Ensemble Strategy | Use of multiple LLMs to produce varied responses, and then ranking them to select the best | 5 LLMs (Parallel usage) | 1 LLM (Parallel usage) |
Analysis and Reporting | Number of monthly tokens per user | 100,000 tokens | 1,000,000 tokens |
API Access | |||
Access to API | Access to API | ||
User Interface | ChatGPT like interface to query LLMs | ||
Deployment | |||
On-Premise or Cloud | Deploy the entire AGS stack (UI, APIs, runtime library, etc.) in your own environment | ||
Integrations | |||
LangChain Integrations | Access to ZeroTrusted.ai's LangChain tools | ||
Zapier Integrations | Access to ZeroTrusted.ai's Zapier Zaps | ||
Azure Cloud Integrations | Access to ZeroTrusted.ai's app on Azure Apps Marketplace | ||
Improved Accuracy | |||
Improve Accuracy & Reliability | Enhance accuracy by optimizing prompts and validating results. | ||
Provide explanations to results | Explains how and why the result were selected | ||
Prevent Hallucinations | Verify that LLM outputs are not fabricated | ||
Analysis and Reporting | Prompt search analysis and report | ||
Threats & Attack | |||
Identify Injection Attacks | Safeguard against malicious inputs designed to exploit model vulnerabilities | ||
Detect Data Poisoning | Detect poisoned results by identifying outliers found using multiple results generated by LLM ensemble | ||
Discover Ethics and Bias Issues | Detect ethical violations and biases in results | ||
Prevent Jailbreaks | Prevent jailbreaks through custom rule based logic that allow companies to flag/block given keywords | ||
Security & Data Privacy | |||
ztPolicyServer - Secure Gateway as a Service | Secure Gateway as a Service | ||
ztDataPrivacy - End to End Encryption (E2EE) | Check for instances of compliance breaches and sensitive data such as Personally Identifiable Information (PII), Payment Card Industry (PCI) data, etc. Rectify any issues by appropriately sanitizing the data. | ||
