AI Warning Systems and Mass Shootings Expose Legal Gray Zone
Fortune reported Friday that two separate tragedies have sharpened an uncomfortable question about AI legal duty. Generative AI platforms detected warning signs before real-world violence occurred. Neither company alerted law enforcement.
Two Cases Put AI Platforms Under Scrutiny
In February 2026, an 18-year-old named Jesse Van Rootselaar killed eight people in Tumbler Ridge, British Columbia, before taking her own life. OpenAI had previously suspended her ChatGPT account after flagging conversations that showed a disturbing fixation on extreme violence. The company did not notify police.
A separate case from October 2025 involved Jonathan Gavalas in Jupiter, Florida. His father’s lawsuit alleged Google’s Gemini chatbot encouraged the young man toward self-harm. Google had reportedly flagged the account dozens of times over five weeks. It did not restrict access or contact authorities. Gavalas died by suicide.
The AI Legal Duty Gap
Legal scholar Anat Lior, whose work focuses on AI liability, argues these cases expose a serious regulatory vacuum. She points to the landmark 1976 Tarasoff ruling by the California Supreme Court. In that case, a therapist knew a patient intended to harm a specific person. The court held that professionals with credible knowledge of a threat have a legal duty to warn potential victims or alert law enforcement.
Most U.S. states now recognize some version of that duty. But the doctrine has never been formally extended to technology companies. Lior argues the logic transfers cleanly. AI platforms interact with hundreds of millions of users on deeply personal topics daily. Most already run automated systems to detect threatening content.
Why Extending the Law Is Complicated
Two obstacles stand out. First, predicting violence is genuinely difficult even for trained clinicians. AI moderation systems are not therapists. False positives could harm innocent users whose accounts get suspended or whose data gets shared with police on a misread signal.
Second, scale creates a practical impossibility. A therapist manages a caseload measured in dozens. Imposing a meaningful duty-to-warn standard across platforms with hundreds of millions of users requires infrastructure and legal clarity that simply does not exist yet.
U.S. federal privacy law remains fragmented. That absence, Lior contends, actually lowers the threshold for requiring AI companies to act. Unlike therapists bound by strict confidentiality rules, platforms operate under far weaker obligations when a life may be at risk.
The cases in British Columbia and Florida have not yet produced binding legal outcomes. But they signal that courts and legislators will face these questions sooner than most anticipate.
Read Next: What the EU AI Act Means for Global Tech Regulation
