Genon, a generative AI solutions company, said on Oct. 26 it has launched its enterprise AI security solution GenGuard.

GenGuard is a solution that judges and blocks requests entered into AI services and generated responses in line with security policies set by a company. It was developed based on guardrail operations experience from the existing generative AI platform GenOS, and can be used regardless of whether GenOS is adopted.

A core feature is natural-language-based security policy setting. Without additional model retraining, users can enter service-specific policies such as blocking individual stock recommendations, and it controls AI requests and responses according to the criteria. It can respond to harmful content, jailbreak attempts and prompt attacks.

It also supports personal information protection. It detects not only structured information such as resident registration numbers and account numbers but also unstructured personal information such as names and addresses. It can block or mask personal information sent to external AI services, and provides an observation mode to check for false positives before applying policies and an on-premises operating environment.

Along with the launch, Genon also disclosed its self-developed lightweight guardrail model, GenGuard 1.0 Light, and a Korean service policy evaluation dataset, KSPG-Bench. According to the company, its model ranked first with an average score of 84.3 in a comparative evaluation against 5 publicly available guardrail models.

Genon plans to expand adoption of GenGuard, focusing on industries such as finance and the public sector where information protection and internal controls are important.

Myeong Daewoo (명대우), a vice president at Genon, said, "As corporate use of AI becomes full-fledged, security will go beyond a simple risk management tool and become a key factor determining the reliability and competitiveness of AI services," and added, "We will lead the market with AI security technology optimized for the domestic corporate environment."