KUALA LUMPUR, Oct 31 — OpenAI is tightening restrictions on weapons-related information available through ChatGPT to users under 18, alongside new safeguards aimed at detecting and preventing potential violence.
In a statement published yesterday, the company said the expanded rules would limit practical information about weapons and security systems that could be misused, even when a teenager’s questions appear harmless or show no indication of violent intent.
“Our policies already prohibit helping anyone plan or carry out violence and restrict helping teens find or obtain weapons,” it said.
“We’re extending those protections to restrict practical information about weapons and security systems that could be misused to facilitate violence, even when there’s no sign of violent intent.”
OpenAI also announced stricter limits for all users on information about past mass attacks, including detailed accounts of the weapons and equipment used, how they were obtained and comparisons that could amplify perpetrators’ notoriety.
The company said it was developing systems that would automatically terminate conversations when users display signs of fixation on violence or seriously harming others, with repeated incidents potentially resulting in temporary account suspensions.
It also plans to take a more cautious approach when previous conversations indicate warning signs of possible violence, even if subsequent questions appear harmless in isolation.
OpenAI disclosed that it had referred more than 200 cases to law enforcement authorities during the third quarter of 2026 over potential threats of violence, saying such referrals had helped disrupt violent plots.
The company said its age-prediction technology would continue assessing users’ behaviour to identify those potentially under 18 and apply additional protections, while adults incorrectly classified as minors could verify their age to have the restrictions removed.
Parents whose accounts are linked to their teenagers’ ChatGPT accounts may already receive notifications when their children’s accounts are disabled for violating policies on violence, with OpenAI also working on alerts for repeated violent discussions that result in account suspensions.
Separately, OpenAI said it would publish its first Community Safety Report in early 2027, covering the second half of 2026, to explain how it detects, investigates and responds to potentially harmful activity.
The report, which will be updated every six months, will include anonymised examples of violence-related risks, changes to safety measures and relevant data on their effectiveness.
OpenAI said the new safeguards would begin rolling out over the coming weeks as it continues refining its systems to identify threats and prevent misuse.
