Artificially Assisted Gun Violence: Chatbot Risks and Preventative Steps for Responsible AI Companies

Key Findings

• — None of the major platforms reviewed addresses firearms explicitly in its consumer-facing usage policies or terms; weapons language ranges from an explicit ban on conventional-weapons procurement or use (OpenAI) to coverage of biological, chemical and mass-destruction weapons only (xAI).
• — No platform has published a commitment to respond differently to firearm acquisition or use questions from users below the legal purchase age.
• — OpenAI is the only company reviewed with a public commitment to notify law enforcement of an imminent and credible risk to others; the other companies permit emergency disclosure without committing to it.
• — The report names three recurring failure patterns across litigated cases and third-party testing: affirming harmful impulses, providing operational help for violence, and not acting on high-confidence warning signs.
• — Recommendations include refusing specified firearm information, a crisis-response protocol when self-harm or violent intent co-occurs with recent firearm discussion, human review of such conversations, and recurring transparency reporting.

Methodology Notes

Desk review of company usage policies, terms, safety statements, court filings and press-obtained chat logs, with policy status as of 18 August 2026. No original model testing; third-party test results are cited, not replicated. Publication timestamp 2026-09-28T17:36Z in the page metadata. Verified by fetching the report page (HTTP 200); authors, date and recommendation text read.

Sources