Designing AI to Stop Crime: A Proactive Safeguarding Framework for Generative AI Platforms
In April 2026, a University of South Florida suspect used ChatGPT to research body disposal, firearm logistics, and evidence concealment before allegedly killing two doctoral students (CBS News, 2026). Months earlier, Florida’s Attorney General launched a criminal investigation into whether ChatGPT advised an FSU campus shooter on weapon selection and optimal timing for maximum casualties. These cases alongside an NFL player’s ChatGPT logs presented as murder evidence and lawsuits alleging chatbots intensified paranoid delusions leading to homicide expose a fundamental design failure. When the USF suspect asked what would happen if a body were placed in a garbage bag, ChatGPT flagged it as “dangerous” but continued engaging when the suspect rephrased. The system treated each query atomically, missing the lethal trajectory across a multi-turn conversation. This paper argues that stopping AI facilitated crime requires a paradigm shift from static content moderation to proactive behavioral safeguarding. Drawing on affordance theory (Volkoff & Strong, 2013) and design science research, we propose a Proactive AI Safeguarding (PAS) framework with three design layers. Layer 1: Behavioral Escalation Detection. Rather than filtering individual prompts, AI platforms should implement session-level behavioral analytics that track query trajectories across time. This moves beyond the static guardrail paradigm toward dynamic, context-aware safety architectures. Layer 2: Tiered Intervention Protocols. Current systems offer a binary response: comply or refuse. The tiered approach addresses the “abrupt refusal” problem where blunt blocking can itself cause secondary harm while maintaining the system’s capacity to intervene meaningfully. Layer 3: AI-Law Enforcement Collaboration Architecture. AI chat logs are already proving more forensically revealing than traditional search histories because conversational exchanges capture reasoning chains, not just keywords. We argue for designing structured collaboration protocols between AI platforms and law enforcement including standardized evidence preservation, transparent disclosure policies, and “glass box” audit trails that enable AI to serve as a crime-fighting ally while protecting civil liberties through governance frameworks like those proposed by the Council on Criminal Justice (2026). This framework contributes to IS research in three ways. First, it extends affordance theory by theorizing dark affordances capabilities that users exploit for harmful purposes and their design-level countermeasures, addressing a gap in how IS scholarship conceptualizes AI misuse. Second, it positions AI not as a passive tool but as a sociotechnical team member in crime ecosystems whose behavior is shaped by design choices, user intent, and regulatory context. Third, it offers actionable design principles for the IS community to engage with one of the most urgent societal challenges: ensuring that AI platforms prevent crime rather than facilitate it. This talk invites debate on the ethical boundaries of proactive AI intervention, the privacy safety tradeoff, and how the IS discipline can lead in designing AI systems that actively protect human life.