There is a Korean artificial intelligence (AI) startup drawing attention from global big tech companies such as OpenAI and Anthropic. It is AIM Intelligence, which provides AI security solutions. Founded last July with CEO Yoo Sang-yoon (28) at the helm, AIM Intelligence builds solutions that serve as both "spear and shield" to protect AI. Yoo said the company is likely to gain more attention as AI's roles diversify and grow more complex. Yoo was interviewed at the company's office in Seocho-gu, Seoul, on the 19th.
◇ Studying "responsible AI"… "AI security is the future"
Yoo, who holds a master's degree from the Department of Electrical and Computer Engineering at Seoul National University, studied responsible AI at school and heard from older alumni that it "felt like reliving the mobile era from 10 years ago." Yoo said, "Technology cycles come every 10 or 15 years, and rather than becoming a part of another company at a time favorable for business, I wanted to start a company and contribute to the new era."
When Yoo said the company would go into the "AI security" business, those around were lukewarm. Advice followed that boosting performance should come before safety. But Yoo countered that "AI you can trust to use will be in the spotlight sooner or later." Yoo recruited the rival who finished runner-up to him at a hackathon as chief technology officer (CTO). Friends who had been building a Blockchain-based "zero trust" game together joined as chief product officer (CPO) and co-founders.
AIM Intelligence has two core products. The first is AIM Red, a Generative AI red team solution. AIM Red probes and attacks AI models and AI-based services to find vulnerabilities. It is a kind of "white hat hacker." Whereas safety had traditionally been evaluated by manually trying a few prompts (commands), AIM Red automatically generates and executes thousands of attack scenarios.
The second is AIM Guard, an AI guardrail service. AI guardrails refer to safety devices configured to prevent AI from taking dangerous actions. AIM Guard provides customized AI solutions that block issues that can occur in AI systems and services diagnosed by AIM Red.
Yoo explained, "AIM Red and AIM Guard operate in tandem. When the red team attacks and finds vulnerabilities, the guardrail is improved to block them, and the attacks are repeated. It is a way to continuously raise AI safety through a kind of 'spear and shield' cycle."
◇ Global big tech is paying attention… new product to launch in the first half of next year
AIM Intelligence's technology has drawn attention from global AI companies. AIM Red by AIM Intelligence conducted an in-depth diagnosis of Anthropic's nonpublic AI model. The goal was to prevent the Anthropic model from outputting CBRN (chemical, biological, radiological and nuclear) manufacturing methods. AIM Red caught the model bypassing existing guardrails to generate weapons manufacturing methods, and Anthropic reflected AIM Intelligence's solution in its company policies.
OpenAI also reached out to AIM Intelligence to receive external verification while preparing its own guardrail system. Kim Kyung-hoon, head of OpenAI Korea, mentioned AIM Intelligence during a presentation at the recent "OpenAI DevDay," introducing it as "a startup developing related technologies so corporations can adopt AI safely."
As "physical AI" is drawing attention as next-generation AI technology, LG Electronics has also teamed up with AIM Intelligence. AIM Intelligence, LG Electronics, and U.S. robotics startup OpenMind are conducting joint research to define dangerous situations for physical AI and build a dataset. AIM Intelligence even spent 60 million won to purchase a robot and set up a test bed.
AIM Intelligence plans to launch a new AI safety solution early next year. It is a service that tracks and controls every action of AI agents. Yoo saw limitations in current guardrail services that monitor individual input and output units in next-generation agent environments performing complex actions. The new product monitors in real time the entire action flow as AI agents browse the internet or interact with external systems, instantly blocking dangerous actions that deviate from user intent.
The company is seeing results just over a year after its founding. Yoo said, "This year's expected revenue is about 1.4 billion won," adding, "In finance, KB Kookmin Card, KB Securities, and Woori Bank, and among telecom operators, KT and LG Uplus adopted AIM Intelligence's AI security solutions."
AIM Intelligence raised a pre-Series A round last August, bringing cumulative investment to 1.85 billion won, and is preparing an investment round early next year.