StarGuard AI reverse proxy scans enterprise LLM traffic for jailbreaks and data leaks
Orion Soft's StarGuard AI acts as a gateway between employees and language models, detecting sensitive data exposure, prompt injections, and jailbreak attempts across ChatGPT, Claude, DeepSeek, and local LLMs.
Orion Soft launched StarGuard AI this week, a security gateway for enterprises running large language models in production. The platform sits between employees and LLM endpoints — ChatGPT, Claude, DeepSeek, GigaChat, YandexGPT, and on-premise models — and scans every request and response for data leaks, jailbreak attempts, prompt injections, and off-policy content.
A joint study by K2 Cybersecurity and Kaspersky Lab found that only 22 percent of companies budget separately for AI security, while 61 percent want to control employee use of third-party AI services but lack the technical or organizational infrastructure to do so. StarGuard AI targets that gap.
What stands out
- 01Reverse proxy architecture. StarGuard AI deploys as a gateway between users, internal systems, and language model APIs. All traffic flows through the platform, which logs and analyzes it in real time.
- 02Threat detection. The system flags sensitive data in prompts (credentials, PII, trade secrets), jailbreak patterns, prompt injection attempts, and content that violates corporate policy.
- 03Audit trail. Every interaction is logged for compliance and forensic review. The platform generates reports on who accessed which models, what they asked, and whether any policy violations occurred.
- 04Multi-model support. StarGuard AI connects to OpenAI, Anthropic, DeepSeek, Yandex, Sber's GigaChat, and self-hosted LLMs. It integrates with OpenWebUI, IDEs, AI agent frameworks, and enterprise platforms.
- 05Target market. Orion Soft positions the product for large enterprises, industrial firms, financial institutions, and government agencies — sectors where data leakage to third-party APIs is a compliance risk.
