South Korean Woman Allegedly Used ChatGPT to Assess Lethality of Drug-and-Alcohol Mixtures Before Two Fatal Motel Poisonings

In Seoul, a woman allegedly used ChatGPT to ask whether mixing sleeping pills or benzodiazepines with alcohol could be fatal before poisoning drinks given to three men. Two men later died in separate motel incidents, and a third survived after losing consciousness. Police reportedly cited her chatbot queries and search history as evidence of intent.

Risk classification

• Primary risk domain: 1 Discrimination & Toxicity
• Primary risk subdomain: 1.2 Exposure to toxic content

The AI system provided information on lethal drug and alcohol combinations, facilitating an illegal act and violence due to a lack of guardrails against assisting in homicidal planning.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The AI system lacked the capability or robust safety filters to detect and block queries aimed at planning a murder.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The harm was caused by a human user who intentionally used ChatGPT to research and execute a lethal poisoning plan.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system involved is ChatGPT, which is a chatbot. The EU AI Act classifies chatbots under Limited Risk, requiring specific transparency obligations to ensure users are aware they are interacting with an AI.

AI system and alleged parties

• AI system: None named
• AI purpose: Chatbot; Question Answering
• Behaviour type: Assistant
• Alleged developer: OpenAI
• Alleged deployer: Kim (suspect in Seoul poisoning case)
• Alleged harmed parties: Three unnamed men in their 20s in Seoul

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Negligible. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Substantial
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Minor
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Physical

Reported: The report explicitly describes physical harm, including two fatalities and one instance of unconsciousness.

Indirectly caused: Two men died from lethal poisoning, and a third man lost consciousness but survived after being given drinks laced with sedatives and alcohol.

Civil rights

Reported: The report explicitly describes events that resulted in the violation of the fundamental right to life of the victims.

Indirectly caused: The lives of two individuals were taken, violating their basic human right to life, facilitated by information obtained from the AI.

People affected

• Occurrences reported: 1
• People reportedly harmed: 3
• People reportedly exposed: 3

Potential causes

Management

• Inadequate Risk Assessment: Management failed to assess risks of users leveraging AI for planning crimes.
• Prioritizing Utility Over Safety: The company focused on factual accuracy over preventing dangerous outcomes.

Technology

• Lack of Guardrails for Safe Info: Chatbot provided factual answers on lethal drug doses without raising alarms.
• No Contextual Intent Detection: The system failed to detect the malicious intent behind factual questions.
• Risk of AI Psychosis Exacerbation: Chatbots can worsen symptoms in users experiencing mental health crises.

Data Inputs

• Unfiltered Factual Training Data: AI training data included precise details on lethal drug combinations.
• Lack of Query Intent Labeling: Input processing lacked labels to flag series of suspicious factual queries.

Human Factors

• User Mental Illness: The user had a mental illness, which can be exacerbated by chatbot use.
• Premeditated Malicious Intent: The user actively sought information to plan and execute multiple murders.

Process and Methods

• Inadequate Guardrail Frameworks: Safety processes only flagged self-harm, ignoring homicidal lines of inquiry.
• No Multi-turn Risk Analysis: The safety methods did not analyze the cumulative risk of sequential chats.

Regulatory Environment

• Absence of Chatbot Safety Laws: A lack of legal standards allowed chatbots to operate without robust safety.
• Lack of Harm Reporting Mandates: No regulatory requirements existed to report user risks or harmful queries.

Information quality

• Classification confidence: High
• Reason for confidence: The report clearly details the role of ChatGPT in providing factual information to a user planning a crime. The facts regarding the murders, the specific questions asked, and the AI's responses are explicitly stated and confirmed by police and OpenAI.

A domestic criminal incident in South Korea where an individual used ChatGPT to research lethal drug-alcohol combinations to commit murder. This incident constitutes a local law enforcement case with negligible national security implications, though it highlights ongoing debates regarding AI guardrails and the retrieval of dual-use or dangerous information.

• Overall national security impact: Negligible
• Scope: Single nation
• Primary target: South Korea
• Alleged perpetrator: Individual

Threat characteristics

• Imminence: Long-term. The immediate incident is resolved with the arrest of the perpetrator, though the systemic policy debate around AI safety guardrails remains an ongoing strategic concern.
• Autonomy: Human-controlled. The AI acted strictly as a passive information retrieval tool assisting a human user who made all decisions and executed the crime.
• Novelty: Evolved capability. Represents an evolution of using standard search engines to research lethal methods, leveraging conversational LLMs to synthesize dangerous information.

Impact by dimension

• Physical security: Negligible. The incident is a localized criminal homicide of three individuals. It poses no threat to critical infrastructure, military systems, or national-scale physical security.
• Information security: Negligible. No association with espionage, classified information compromise, or systematic information warfare campaigns.
• Sovereignty: Negligible. The event is a standard domestic law enforcement matter with no disruption to state authority, elections, or core government operations.
• Economic security: Negligible. No impact on strategic industries, financial markets, supply chains, or national technological competitive advantage.
• Societal stability: Negligible. While a tragic violation of individual rights to life, it represents isolated criminal activity rather than systemic, population-scale social manipulation or civil unrest.