Is AI becoming a silent partner in US Mass shootings? Report details dangerous gaps

A student asks an AI chatbot when a college building will be busiest. He follows with questions about guns. Each question may look harmless when viewed alone. Together, they could reveal something far more dangerous.

A new Everytown for Gun Safety report argues that chatbot companies are missing patterns that could help prevent US mass shootings. The report examines court filings and independent research. It also reviews the safety policies of major AI companies, as per a report by USA Today.

Everytown is a gun-safety advocacy group. Its conclusions are not court findings. Yet the cases it highlights have raised an urgent question: When should a chatbot stop answering and start sounding an alarm?

FSU Lawsuit Puts AI and US Mass Shootings Under Scrutiny

Phoenix Ikner has been charged in the April 2025 shooting at Florida State University. Two people were killed. Several others were wounded. Ikner has pleaded not guilty.

A lawsuit filed by the family of victim Tiru Chabba alleges Ikner used ChatGPT while preparing for the attack.

According to the complaint, Ikner asked about busy periods at the FSU student union. He also discussed firearms and school shootings. ChatGPT allegedly supplied factual information about his weapons.

The lawsuit claims the conversations revealed enough warning signs to justify intervention. OpenAI rejects the accusation.

An OpenAI spokesperson told Reuters that ChatGPT did not promote illegal activity. The company said its responses contained information already available from public sources.

Researchers Say Chatbot Guardrails Can Be Bypassed

Everytown’s report goes beyond a single case. Researchers from the Center for Countering Digital Hate tested 10 popular chatbot platforms. They posed as teenagers planning violent attacks.

Eight platforms provided actionable assistance during at least some conversations. Anthropic’s Claude and Snapchat’s My AI refused harmful requests more consistently, according to the study.

Researchers said several bots offered firearm advice. Some also supplied tactical guidance.

The findings came from simulated conversations. They do not prove a chatbot will always respond in the same way. Yet they show how safety systems can fail when dangerous questions are disguised as research or fiction.

OpenAI Says Threats Can Trigger Human Review

OpenAI says ChatGPT is trained to reject requests that could meaningfully enable violence.

Automated systems monitor conversations for concerning activity. Flagged chats can be reviewed by trained staff. Accounts may be disabled when serious policy violations are found.

OpenAI also says it may contact police when reviewers identify an imminent and credible threat to others. The company acknowledges that warning signs may emerge slowly across a long conversation. No court has ruled that ChatGPT caused the FSU shooting. The lawsuit remains contested.

What Everytown Wants AI Companies to Change

Everytown wants chatbots to stop providing unsafe firearm information once violent intent appears.

A proposed crisis system would keep restrictions active throughout the conversation. A user could not remove the safeguards by claiming the questions were for a school assignment.

High-risk conversations involving guns would also receive human review. Serious threats could then be referred to law enforcement.

Everytown also wants companies to detect users who create new accounts after being banned. Clear public safety policies would explain when private conversations may be reviewed.

AI may not create violent intent. A poorly designed chatbot could still strengthen it or help turn vague thoughts into a workable plan.

FAQs

What does the Everytown report say about AI?

The report says chatbots may provide dangerous information when users show signs of violent intent or suicidal thinking.

Did ChatGPT cause the Florida State University shooting?

A victim’s family alleges ChatGPT assisted the accused shooter. OpenAI denies responsibility. The lawsuit has not been decided.