StarGuard AI: Why Enterprise LLMs Need a Firewall, and How Orion Soft Is Building One

In July 2026, Orion Soft announced StarGuard AI, a security gateway (firewall) for large language models in corporate environments. The product sits between users and models and controls how employees, developers and internal services access on-premise and cloud LLMs.

Why businesses suddenly need a firewall for LLMs

Over the past year, large language models in the enterprise have gone from isolated experiments to being built into daily workflows, and that shift has created a new risk surface. While a model was one team's toy, there was nothing to protect. Once an LLM became part of the logic of internal products, every request to it turned into a channel through which data can leak or a malicious instruction can arrive.

According to Orion Soft, four ways of using LLMs have grown at companies at the same time over the past year:

• IT teams are deploying local (on-premise) models
• Business users work through cloud services
• Developers connect AI agents directly in their IDEs
• Internal products use LLMs as part of their business logic

Each of these four channels runs under its own access rules. Without a single control point, the security team cannot see who is sending what to a model, or where.

What the StarGuard AI gateway protects

StarGuard AI works as a single control point for all requests to LLMs, much like a network firewall, but for traffic going to models. The idea is to give corporate security one manageable gateway instead of dozens of disconnected integrations with cloud and local services.

The typical threats that gateways like this are built to address fall into three areas: leakage of sensitive data when requests are sent to external cloud LLMs; prompt injection, where malicious instructions are mixed into user input; and "shadow AI," where employees use models in violation of company policy. A single gateway makes it possible to see and filter this traffic in one place.

"This is exactly the problem we are building

StarGuard AI for: a security gateway for large language models," said Nikita Vekesser, product manager for AI infrastructure at Orion Soft.

In Orion Soft's blog on Habr, Vekesser notes that the transition from experiments to integration was "short but noticeable": the technology was adopted faster than companies could build a security perimeter around it. The gateway targets that gap between the speed of LLM adoption and the maturity of protection.

How an LLM firewall differs from a network firewall

An LLM firewall controls not network packets but the meaning of requests and model responses. It inspects the content of prompts and generations, not just IP addresses and ports. A conventional network firewall cannot tell that a request to an LLM contains an attempt to extract a corporate secret or bypass an instruction. A specialized gateway is designed for precisely these four scenarios of corporate model use.

This reflects a broader shift: a separate layer of infrastructure is beginning to take shape around language models, covering access control, request auditing, and input and output filtering. That layer used to be built around networks, databases and APIs. Now it is LLMs' turn.

What this means

AI infrastructure is entering a phase in which a separate security perimeter is growing around language models, just as one once grew around networks and APIs. The arrival of specialized LLM firewalls such as StarGuard AI signals that corporate use of AI is ceasing to be an experiment and is starting to require the same controls as any other critical IT system.