Constellation Gate AI Launched: Security Gateway Boasts Top-Ranked Prompt Injection Protection Test Results
A NASDAQ-listed company launches an AI security solution that integrates with existing systems without code changes, showcasing standard test results with an F1 score of 97.4%.
AIAI Holdings (NASDAQ: $AIAI) has launched a new product in the AI security market through its subsidiary, Constellation Network, named "Constellation Gate AI." It is a security and auditing gateway specifically for artificial intelligence applications. At its core, Gate AI acts as an intermediary between user applications and large language models (LLMs) to prevent attacks known as Prompt Injection (covert command attacks), a critical vulnerability where malicious actors can input special commands to control or steal data from AI. A key feature of this product is that developers can integrate it directly into existing AI workflows without needing to modify legacy code. It also comes with an Audit Trail system that meticulously records every request and response, and a single dashboard for tracking all usage and costs. To prove its effectiveness, the team released test results from a technical report, stating that Gate AI was evaluated against 16 public benchmark datasets for Prompt Injection protection. The results show that Gate AI ranked first in 8 datasets and placed within the top three in another 4 datasets, achieving a high F1 score (a measure of model accuracy) of 97.4% with a False-Positive-Rate of only 1%. In terms of speed, the end-to-end response median was 53 milliseconds, and the average was 104 milliseconds. Ben Jorgensen, CEO of Constellation, stated, "Developers are only one prompt injection away from losing business effectiveness," emphasizing that Gate AI is designed to block attacks before they reach the AI model, while logging all activities for retrospective auditing. Constellation Gate AI will be available in two forms: one where customers connect their existing model provider's API keys to utilize security and auditing features, and a "Pay-as-you-go" model where users can directly access both commercial and open-source AI models via Gate AI without needing to manage relationships with multiple providers.
Prompt Injection attacks are a critical vulnerability in current AI applications. This tool offers an an alternative for Thai developers and organizations looking to enhance the security of their AI systems without modifying existing code.