The Journey Toward Trustworthy AI: Introducing LG AI Research's 'AI Ethics Seminar'

LG AI Research believes that responsible development and use of AI are as important as technological innovation itself, and operates an 'AI Ethics Seminar' to put AI Ethics into practice across the organization.

Recently, AI is evolving beyond a simple conversational partner into Agentic AI, which sets its own goals and uses tools. This signifies an era of autonomous colleagues that move beyond the existing 'input-output' model and operate through a 'perception-reasoning-action-feedback' loop.

However, as AI is granted not just the 'authority to speak' but the 'authority to act,' new security challenges have emerged. While security in the era of existing chatbots focused on 'content safety,' such as blocking hate speech, in the era of agents, Action Safety—ensuring that AI's actions do not cause unintended harm—and Goal Integrity—ensuring that AI does not deviate from the user's intent—are essential.

In particular, attention must be paid to the 'Agency-Security Paradox,' which arises when the LLM's statistical reasoning capability is combined with system control authority. As AI becomes capable of performing real actions such as modifying accounts or deleting files, a single reasoning error or memory manipulation can directly lead to an actual system security incident.