New Everytown White Paper Documents Growing Risk of AI Chatbot-Assisted Gun Violence

NEW YORK — Today, Everytown for Gun Safety Support Fund released Artificially Assisted Gun Violence: Chatbot Risks and Preventative Steps for Responsible AI Companies, a new white paper on how consumer AI chatbots have intersected with real-world gun violence. The first paper to explicitly tackle the intersection of gun violence and AI, it surveys five U.S. companies that build frontier AI models and run their own consumer chatbots: OpenAI (ChatGPT), Google (Gemini), Meta (Meta AI), Anthropic (Claude), and xAI (Grok). 

The paper documents patterns of dangerous chatbot behavior in court filings, public reporting, and independent testing, and compares what each company has publicly committed to on safety guidelines around weapons, self-harm, and law enforcement escalation. It closes with recommendations that AI companies can adopt now, without waiting for courts or legislatures to require them. Everytown’s central conclusion is that chatbot behavior is the result of deliberate company choices on design and safety, and companies must take stronger steps to prevent the risk of their products facilitating gun violence. 

Artificially Assisted Gun Violence is available here.

Key Risk Areas:

• Suicide “coaching”: Lawsuits allege that chatbots affirmed young users’ suicidal thoughts instead of intervening, in one case while the user described holding a loaded gun.
• Attack planning: According to lawsuits and news reports, the man who opened fire at Florida State University in April 2025 asked ChatGPT how many victims a school shooting needs to make national news. It answered.
• Missed warnings: OpenAI allegedly flagged the Tumbler Ridge, B.C. school shooter for “gun violence activity” eight months before the attack. It then deactivated the account without alerting police. The government of British Columbia is now suing OpenAI.
• Industry-wide failures: In independent testing, 8 of 10 popular chatbots helped researchers posing as teens plan violent attacks.
• No firearm policies: None of the five major AI companies explicitly addresses firearms in its consumer usage policies.

Recommendations: 

• Refuse dangerous requests: Block information on illegal firearm acquisition and modification, including buying a gun without a background check, converting a firearm to fully automatic, and building untraceable guns.
• Trigger a crisis response: When a user pairs violent or suicidal intent with interest in firearms, switch the chatbot into a crisis protocol that can’t be undone by reframing or roleplay.
• Put humans in the loop: Require human review of high-risk conversations involving firearms, and refer serious threats to law enforcement or mobile mental health crisis teams.
• Be transparent: Publish firearm safety policies in one place and audit them for the life of the product.
• Promote gun safety: Equip chatbots to share accurate information on secure storage, Extreme Risk laws, and crisis resources.

To speak with an expert from the Everytown for Gun Safety research team, please contact [email protected].

###