OpenAI has introduced tighter ChatGPT safety restrictions to block teenage accounts from receiving practical details about weapons and security systems, even when queries show no violent intent. The measures, announced on Friday, also restrict historical details about mass casualty events for all users across the platform.
Key points
- ChatGPT will block teenage accounts from practical queries about weapons, ammunition and security setups, regardless of context.
- Accounts showing persistent fixation on violence or harm face automated chat terminations and temporary account suspensions.
- Historical reconstructions of mass attacks, weapon acquisition and casualty comparisons will be withheld from all users.
- OpenAI referred more than 200 cases to law enforcement agencies during the third quarter of 2026.
Stricter boundaries on weapons and security
The updated rules prevent ChatGPT from supplying teenage users with actionable instructions regarding weapons, explosives, or how to circumvent home and institutional security alarms. OpenAI stated in a blog post that these limits will apply to teenage accounts automatically, even when an enquiry appears purely educational or benign.
For adult and teenage users alike, the system will also reduce granular detail about previous mass attacks. The model will no longer reconstruct past attacks, detail how perpetrators acquired their equipment, or supply close comparisons of the harm caused. OpenAI confirmed that users can still enquire about warning signs, psychological impacts, and the broad historical consequences for victims and communities.
Automated terminations and account suspensions
OpenAI is expanding the technical triggers that allow the model to terminate active conversations. When ChatGPT detects patterns indicating that an individual is fixated on violent actions or severe physical harm toward others, it will cut off the interaction immediately.
Repeated session terminations for violent fixation may result in a temporary suspension of account access. OpenAI noted that it made more than 200 referrals to law enforcement authorities in the third quarter of 2026 where imminent threats of physical harm were detected.
The company also widened its age prediction models, which evaluate user behaviour to identify minors under 18. Users identified as minors are routed into restricted teen profiles; adults placed in this mode by mistake must submit identification documents through third-party identity platform Persona.
Parental notifications and legal pressure
The adjustments build on teen protections introduced in August. Since July, parents who link their accounts to a teenager receive notifications if an account is permanently shut down because of violent content. OpenAI is now building an additional alert system to notify parents whenever an account is paused for repeated discussions of violence.
These safety updates arrive as generative AI providers face rising scrutiny over youth safety. Common Sense Media warned earlier this month that the teen experience remained inadequate for underage users. Eight US states have filed legal actions against OpenAI alleging harm to young users, while the state of Texas has brought a lawsuit against OpenAI and chief executive Sam Altman.
OpenAI plans to publish its inaugural Community Safety Report in early 2027, covering the final six months of 2026, with successive reports scheduled every six months to detail moderation actions and anonymised safety data.
