Unveiling GuArdIan Enterprise Sandbox: A CXO’s Introduction to Safe, Secure and Compliant Gen AI Deployments

AIShield is thrilled to announce the launch of its GuArdIan Enterprise Sandbox designed for IT managers responsible for ensuring adherence to safety, security and compliance in Gen AI based applications of an organization. The Sandbox is a containerized, easy to deploy version with core functionalities of the GuArdIan Solution – AIShield’ s state-of-the-art cybersecurity middleware designed for GenAI application security of enterprises. The GuArdIan Enterprise Sandbox allows admin users to configure and test for security integrity checks, content access controls, content analysis, content safety, privacy protection along with user management. The multimodal Sandbox is designed to be LLM agnostic and to seamlessly integrate its observability layer with the existing SIEM tools in the organization. The GuArdIan Enterprise Sandbox is a testament to AIShield’ s commitment to its customers and partners’ needs to experiment ,understand and manage the nuances of safety, security and privacy risks of their Gen AI applications, towards an informed decision making in their PoC to Production journey.

Since the rapid rise of Generative AI (Gen AI) in early 2023, companies worldwide have been actively exploring its capabilities across various functions and business contexts. In fact, Gartner expects that by 2024, nearly 60% organizations plan to move from PoCs to production. However, a stark reality of managing the safety and security risks while transitioning from controlled PoC environments to production scale implementations presents a significant hurdle in the widespread adoption of this technology.

More than 84% of enterprises, in a 2023 IBM study, see cybersecurity risks as the topmost roadblock to Gen AI adoption. Meanwhile, Gartner reports that nearly 30% of enterprises deploying Gen AI had AI security breaches, majority of which were related to data compromise (IP and PII) by internal and external parties, followed by malicious attack on their AI infrastructure. Additionally, key concerns around ethical usage, appropriateness of content and adherence with various upcoming regulations pose more challenges.

Moreover, a vast array of Gen AI tools can overwhelm decision-makers, particularly when considering critical aspects such as safety, security, and privacy. This strain on project teams and security officers may hinder businesses from fully leveraging the technology's benefits at scale.

In early 2023, we introduced the AIShield GuArdIan to the world. GuArdIan, a key offering of our Unified AI Security Platform SecureAIx is a state-of-the-art cybersecurity middleware developed to safeguard multimodal Gen AI applications against key concerns like IP and PII data leaks, reputational and legal risks associated with access to inappropriate content and security risks like prompt injection, jailbreak, and evasion attacks, among others.

Today, we are excited to launch our GuArdIan Enterprise Sandbox – a containerized, easy to deploy version of the AIShield GuArdIan that offers enterprises an opportunity to experience the core features of AIShield GuArdIan solution in a controlled environment of their own. While GuArdIan helps organizations mitigate safety, security and privacy risks for production level Gen AI deployments, the GuArdIan Enterprise Sandbox is designed to seamlessly integrate into their PoCs for wholesome experimentation without fear of security risks.

The GuArdIan Enterprise Sandbox is predominantly aimed to provide the Internal Security Teams a tool to ensure the security and safety of Gen AI driven automation & implementation into the organization’s systems and processes.

The Sandbox is a testament of AIShield’ s commitment to its customers and partners’ needs to experiment and understand the nuances of safety, security and privacy risks of their Gen AI innovations for an informed decision making in their PoC to Production journey.

The Sandbox, with its simple and easy to use GUI interface, equips the IT Security teams (Admin users) with a tool to easily configure security and safety controls, test their effectiveness in real time as well as observe the usage and results for further reporting and analysis. The three key modules of the Enterprise Sandbox are:

1. Configuration Module for real-time and straightforward Policy Configuration

The configuration module allows the admin users to configure organization and content related policies along with security controls from the generally available feature set for multiple applications and end users on a simple UI screen. The extensive yet non-exhaustive policies include

a. Security integrity checks

i. Detects manipulation attempts with prompt injection and jailbreak checks.

ii. Eliminates invisible text

iii. Safeguards sensitive information - secret detection and redaction

b. Content access controls

i. Blocks competitor mentions

ii. Substring filtering

iii. Topic filtering and whitelisting

c. Content analysis-based controls

i. Detects Programming code – prevents code injection threats

ii. Identifies content language

iii. Regex patterns for flexible and powerful content filtering

d. Content safety controls

i. Filters toxic and harmful language

ii. Blocks profanity

iii. Detects biases

e. Privacy protection

i. Data anonymization – regulatory compliance and trust

ii. Detects PII for user privacy and data security

f. User Management

All of this with a Single Pan of control to support multiple application specific security control configuration to support diverse level of protection per application in accordance with organization policy and need.

2. Testing Module for ensuring the effectiveness of the controls and the Sandbox.

Admin users can, in real time, assess the effectiveness of the controls set in place with user-generated input and LLM provided output test prompts across each application setting.

3. Observability and Reporting Module

The Observability module is a comprehensive dashboard, offering a holistic view of end-users' interactions with the Gen AI application. This module provides invaluable insights into the effectiveness of deployed policies by presenting a detailed breakdown of blocked and allowed queries or outputs. Administrators gain visibility into the real-time impact of configured policies, enabling them to assess the system's performance and efficacy. Additionally, the observability module offers essential metrics such as the number of configured policies, the most violated policies, and other key performance indicators, empowering stakeholders to make informed decisions and optimize their AI security strategies effectively.

1. Large Language Models (LLMs): The Guardian Sandbox is LLM agnostic. Widely used LLMs such as Azure OpenAI, AWS Llama 2, AWS Claude and more can easily be configured with a simple UI, providing flexibility to businesses for securing their applications built on any LLM.

2. SIEM tools: The Sandbox can be smoothly integrated with the most popular SIEM tools like Azure Sentinel and Splunk using our applications available in respective marketplaces.

3. Deployment environment: The Sandbox mirrors the flexibility of GuArdIan in being an on-premise solution and can be deployed either in customer’s cloud environment or their data centers as a single container (can be scaled with Kubernetes, if needed) in minutes.

• Enhanced Multimodal User Experience: The Sandbox offers seamless integrations with a variety of Large Language Models (LLMs) and Security Information and Event Management (SIEM) systems, coupled with the simplicity of policy configurations through an intuitive interface. This ensures a smooth implementation of controls and policies.
• Easy to deploy within minutes: The Sandbox is single docker container version of AIShield GuArdIan and can be swiftly and effortlessly deployed in any target environment within minutes.
• Ease of configuring and testing controls for multiple apps and users: The Sandbox allows admin users to configure and enforce controls for multiple Gen AI applications with a single pan of control.
• Cost effective for Gen AI PoCs: The Sandbox is efficiently priced in order to enable enterprises to experiment with security controls in conjunction with their Gen AI applications without budgetary constraints.
• Actionable insights for Gen AI Security strategies: Security teams can leverage the reports and dashboards available in the Sandbox to drive evidence based AI security strategy for the organization.
• Faster and effective Proof of Value: Faster, cheaper and efficient experimentation without full scale deployment leading to faster proof of value that can be articulated to the leadership for decision making to move from PoC to production.
• Alignment with Industry leading standards: The Enterprise Sandbox, like AIShield GuArdIan, aligns with the industry recognized standards like MITRE ATLAS, OWASP and NIST AI RMF.
• Regulatory Compliance: The Sandbox is designed to provide evidences and reports that fulfill the upcoming regulatory requirements of the EU AI Act and the US Executive Order on AI Security.

Dive into the future of AI security with the GuArdIan Enterprise Sandbox.

To get a sneak peek into our Sandbox demo video, click here.

Get your Sandbox access now and lead your organization into this new era of AI with confidence.

It's more than a launch; it's our commitment to ensuring every Gen AI innovation is secure and responsible. We are just getting started!