This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A Florida woman was wrongfully jailed for two days after her ex-boyfriend allegedly used AI-generated text messages to falsely report a bond violation. The incident highlights the growing challenge for the criminal justice system in verifying digital evidence. Following eight months of legal proceedings, the charges were dropped and the victim was later acquitted of the original battery charge. The case has prompted advocacy for new legal standards regarding AI-generated evidence.
A Florida nurse, Melissa Sims, was reportedly jailed for two days for purportedly violating a bond/no-contact order. Reportedly, her ex-boyfriend produced purportedly AI-generated text messages or a "text receipt" attributed to her.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
The ex-boyfriend used AI to generate fake text messages to impersonate the victim and manipulate the legal system, targeting her for personal vendetta and causing her wrongful arrest.
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 harm was caused by the intentional actions of a human (the ex-boyfriend) who used an AI tool to generate fake evidence to frame the victim.
Added categories peak at 2: Minor (Psychological Harm, Indirect)
An individual used AI-generated text messages to falsely frame a victim for a bond violation, leading to a wrongful arrest and two-day detention. While highlighting emerging challenges for digital evidence verification in local judicial systems, the incident represents a localized domestic dispute with negligible national security implications.
The AI system was used by the ex-boyfriend to perform a specific function, namely generating fake text messages, acting as a deterministic tool based on direct inputs.
Risk Level 3. Limited Risk: The system is used to generate AI-generated content such as deepfakes or synthetic text, which requires transparency but does not fall under prohibited or high-risk categories by itself, though its malicious use led to harm.
The AI was used to generate fake text messages that mimicked the victim's writing style to frame her for a bond violation.
The report clearly outlines the sequence of events, the parties involved, the nature of the AI-generated content (texts), and the specific harms (wrongful arrest and jail time). There is little ambiguity about what transpired.
None
Technology
Advanced AI Generation
AI text generation is highly realistic and difficult to spot.
Unreliable AI Detection
Detection tools returned inconsistent results ranging from 1% to 62%.
Data Inputs
Unverified Digital Evidence
AI-generated texts were submitted as evidence without any verification.
Human Factors
Malicious Use by Actor
Ex-boyfriend created fake AI texts to falsely accuse her.
Lack of Police Due Diligence
Police and prosecutors accepted the fake texts without verification.
Process and Methods
Trust-First Evidence Standard
The current standard trusts digital evidence unless proven otherwise.
Inadequate Verification Process
The justice system lacks robust processes to verify AI evidence.
Management
Lack of Prosecutorial Oversight
Prosecutorial management failed to enforce evidence verification.
Regulatory Environment
Lack of AI Evidence Standards
There were no established legal standards for validating AI evidence.
Absence of Forgery Laws
A lack of digital forgery laws failed to deter the deepfake creation.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
