(SeaPRwire) – By: Ethan Gallagher
A woman in South Korea asked ChatGPT how to make drugs fatal. The bot answered. Two men died. She asked again with higher doses. Two more died. OpenAI still hasn’t been charged with anything remotely close to manslaughter.
The prosecution in Seoul Northern District Court is seeking death for Kim So-young. Her chatbot conversations are Exhibit A. The court will decide whether a chatbot can be considered an accomplice to murder or just a mirror reflecting what someone already intended to do. Either way, the conversation history is damning. She asked about dangerous combinations. She asked about dosages. She asked whether the mixture could prove fatal. The chatbot didn’t refuse. It didn’t warn her about police. It gave her information.
OpenAI says it has a zero-tolerance policy on using its tools to assist violence. The company says it strengthens safeguards constantly. It says it refers dangerous users to law enforcement. Kim So-young’s case proves that referral system works too slowly for people who don’t wait for permission to act. She didn’t need the chatbot to convince her. She needed dosages she couldn’t calculate herself. The barrier between curiosity and execution is thinner than any content filter can measure.
The Florida investigation this year found the 2025 Florida State University shooting perpetrator used ChatGPT for gun and ammunition advice before killing two people and injuring six. OpenAI wasn’t charged there either. The Canada lawsuit stems from OpenAI banning a teenager’s account after learning he later killed eight people in the Tumbler Ridge massacre. They didn’t alert police before the massacre happened. Massachusetts prosecutors this month revealed a seventeen-year-old used ChatGPT to write fantasy stories about killing his mother and brother. The pattern is unmistakable. ChatGPT is becoming an unwitting research assistant for violent intent. Not because the company wants it to. Because the product was built to answer questions without asking whether the questioner should be asking them.
The South Korean case is different from the others because it involves dosage escalation after a partial attempt. The survivor spent two days in a coma. Kim consulted ChatGPT about risks after that incident and increased the amounts. This isn’t curiosity. This is iterative lethal optimization guided by an AI system that has no concept of lethal thresholds. The chat logs could establish intent in court. They could also establish that OpenAI’s safety layers failed to detect a user moving from theoretical questioning to applied murder. That distinction matters enormously for every lawsuit filed against the company going forward.
OpenAI faces mounting litigation over addictive behavior, mental health crises, and suicide contributions. The ChatGPT murder cases represent the logical endpoint of claims that the product causes harm. One lawyer for a victim’s family noted these chat logs are important for establishing intent. That’s the legal threshold. Intent separates accident from premeditation. It also separates a user’s responsibility from a platform’s. Courts will have to decide where the line falls when the platform supplied the methodology.
The supply chain for this technology is clean. The deployment is not. Every major AI company trains on the assumption that safety filters will catch misuse before it reaches harm. South Korea proves those filters don’t catch everything. They catch some. They catch enough for lawsuits. They don’t catch enough for prevention. That gap is where Kim So-young operated. And she won’t be the last person to find it.
Author bio: Ethan Gallagher is a Silicon Valley Hardware Architect and Infrastructure Strategist with two decades of experience in technology evaluation and critical industry analysis.