ITCEN Group said on the 1st that it released AgentGo Guard, an integrated AI guardrail security solution that supports corporations and public institutions in adopting generative artificial intelligence (AI) and building an agent ecosystem.
AgentGo Guard is a solution that controls security threats that can occur in Generative AI and AI agents, such as prompt injection, sensitive information leakage, and jailbreak attempts. Developed as a standalone platform, it can be linked not only with ITCEN Group's multi-AI agent management platform AgentGo 2026 but also with AI agents and retrieval-augmented generation (RAG) environments that corporations have already built.
AgentGo Guard features multilayered defense and adversarial attack protection. It runs a multistage detection system from blacklist and regex filtering to natural language-based context analysis and large language model (LLM)-based guardrails, and it can identify and block in real time at the session level "echo chamber" attacks that gradually neutralize defenses through repeated conversations, as well as intelligent evasion and jailbreak attempts.
It also comes with data loss prevention (DLP) capabilities. It supports the Korea Financial Telecommunications and Clearings Institute CMS account number formats (17 types) and the Ministry of Land, Infrastructure and Transport vehicle number regex, and through a Natural Language Processing (NLP) model compliant with Telecommunications Technology Association (TTA) major categories (15 types), the U.S. National Institute of Standards and Technology (NIST), and the EU General Data Protection Regulation (GDPR), it can de-identify (mask) sensitive information such as Korean names, addresses, and resident registration numbers.
Before corporations apply security policies, they can verify security policies and attack scenarios in advance in a virtual sandbox through a pre-simulator. This allows precise pre-adjustment of the optimal policy threshold between user inconvenience due to excessive blocking (false positives) and security gaps (false negatives).
All call logs are stored after AES-256 encryption and cross-site scripting (XSS) prevention processing. Administrators are required to enter a reason when viewing logs, strengthening audit tracking. If anomalies exceed a set threshold, an email is sent to the person in charge.
AgentGo Guard can be deployed and run not only on public networks such as the cloud but also in containerized Docker and Kubernetes environments on closed networks. It also integrates with third-party agent platforms, allowing corporations already operating AI agents to add a security layer without separate system overhauls.
Jeong Seung-mo, head of the AI Innovation Center at ITCEN GLOBAL, said, "AgentGo Guard is the core defensive line of enterprise AI that preemptively controls unethical or hateful expressions by AI, prompt injection, confidential leaks, and false information caused by hallucinations," adding, "We will actively support corporations in maximizing AI productivity safely and without risk."