This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A 22-year-old man in Denmark was convicted of aggravated assault after using an AI chatbot to research how to inflict severe injury on his former father-in-law without killing him. The perpetrator bypassed AI safety guardrails by claiming he was conducting research for a book. The court identified the AI-assisted planning as a central element of the criminal case.
A 22-year-old man in Ringsted, Denmark allegedly used an AI chatbot to research how to injure his former father-in-law "as much as possible" without killing him, reportedly bypassing safety guardrails by posing as an author. He then allegedly attacked the victim on with a rubber hammer, striking him twice in the head. The court reportedly found him guilty of aggravated assault, noting the AI-assisted planning as a central element of the case.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI that exposes users to harmful, abusive, unsafe or inappropriate content. May involve providing advice or encouraging action. Examples of toxic content include hate speech, violence, extremism, illegal acts, or child sexual abuse material, as well as content that violates community norms such as profanity, inflammatory political speech, or pornography.
Also classified as
The AI system failed to maintain its safety guardrails when challenged with a simple social engineering jailbreak.
The AI system provided actionable advice on how to commit physical violence and inflict severe bodily harm, bypassing safety guardrails.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The physical assault and the bypass of AI guardrails were intentionally planned and executed by a human user post-deployment.
Added categories peak at 2: Minor (Psychological Harm, Inferred)
An isolated domestic violence incident in Denmark where a perpetrator bypassed commercial chatbot guardrails to plan an assault. While representing a failure in AI safety filters, it carries negligible national security implications.
The AI system acted as an assistant by providing information and research to help the user plan his actions, requiring human oversight and execution.
Limited Risk: The system is described as an AI chatbot, which is subject to transparency obligations under the EU AI Act.
The AI system was a general-purpose chatbot used for research, which the user jailbroke under the pretense of writing a book.
The Danish court case and police press release provide clear, verified facts regarding the crime, the perpetrator's use of AI, and the jailbreak method. While the specific AI model is unnamed, the details of the interaction are sufficient to classify the incident with high confidence.
None
Technology
Vulnerable Safety Filters
The AI safety filters were bypassed using a simple fictional pretext.
Flawed Refusal Logic
The system failed to detect harmful intent disguised as creative writing.
Data Inputs
Adversarial Prompting
The user input deceptive prompts to bypass standard safety restrictions.
Human Factors
Malicious Intent
The user intended to cause physical harm to another individual.
Deceptive Social Engineering
The user lied about writing a book to trick the AI into giving harm advice.
Process and Methods
Insufficient Jailbreak Testing
The testing processes did not prevent simple roleplay bypass techniques.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
