OpenAI has announced new restrictions on how ChatGPT responds to questions about firearms and violence. The changes apply to all users but include broader protections for teenagers and additional measures to identify conversations that could signal a real threat of violence.
The company presented the measures on Friday, October 30, as part of a review of its safety procedures. OpenAI said it plans to improve its analysis of conversational context and expand communication with authorities in situations considered threatening.
The company also announced a Community Safety Report. Its first edition is expected in early 2027 and will include data from the second half of 2026, covering the identification of threats, responses to concerning behavior and the effectiveness of prevention measures.
OpenAI said it made more than 200 communications to law enforcement during the third quarter of 2026. These referrals involved cases assessed as potentially serious threats. The company added that specialists in mental health and behavior can help evaluate complex situations, alongside consideration of family members and public agencies.
The restrictions will also affect adult users. ChatGPT will provide less detailed responses about past mass attacks and the people responsible for them. The goal is to avoid combining publicly available information in ways that could facilitate the reproduction of violent acts or increase the notoriety of their perpetrators.
The system is expected to limit detailed reconstructions involving weapons and equipment, how they were obtained and comparisons of the scale of the damage. OpenAI said such information could turn attackers into prominent figures and provide references that encourage imitation.
Users will still be able to ask about what happened, the effects on victims and communities, and warning signs that may help identify dangerous situations. The restrictions will focus mainly on operational details that could assist in planning or carrying out a new attack.
ChatGPT will also place greater weight on the context of recent interactions. A question that appears harmless on its own may take on a different meaning when preceded by messages suggesting violent intent. If the system detects signs that someone is fixated on violence or on seriously injuring another person, it may automatically end the conversation. Repeated terminations for the same reason could lead to a temporary account suspension.
Users under 18 will face additional restrictions on weapons-related information. Even when a question appears harmless and contains no clear indication of violent intent, ChatGPT will limit access to certain practical details that could be misused.
Existing rules already prohibited guidance for planning or carrying out violent acts and restricted assistance with finding or obtaining weapons for teenagers. The updated protections will also limit practical information about weapons and safety systems when it could facilitate violence, even without an explicit statement of intent.
Teenagers will continue to be able to use ChatGPT for violence-prevention information, support and ways to contact a trusted adult. OpenAI said the distinction is between content that helps users understand or avoid dangerous situations and practical instructions that could facilitate violent actions.
To apply these protections, OpenAI uses an age-prediction system that considers several signals, including usage patterns. The company said it has improved the tool to assess new signals continuously and direct users who may be under 18 to an experience with additional safeguards.
Adults incorrectly classified as teenagers will be able to confirm that they are over 18 through Persona, OpenAI’s age-verification partner. Once verification is complete, the additional protections for minors can be removed.
The company is also expanding parental controls, which allow a guardian’s account to be linked to a teenager’s account and certain safety settings to be managed. OpenAI already provides alerts on linked accounts when trained reviewers identify serious concerns involving self-harm or behavior associated with eating disorders.
In July 2026, the company expanded those alerts to include cases in which a teenager’s account was deactivated for violating violence-related policies. OpenAI now plans to notify guardians when an account is temporarily suspended after repeated conversations about violence.
The new protections are expected to begin rolling out in the coming weeks. OpenAI said it will continue adjusting its systems to identify warning signs and determine when to limit responses or end conversations, while emphasizing that digital tools should work alongside families, professionals and authorities.
Sources: Olhar Digital.