OpenAI today announced a package of safety measures centered on a new Community Safety Report and immediate updates to how ChatGPT handles violent content and teen accounts. The company said the recurring report will explain how its systems detect, investigate and mitigate risky behavior, while the product changes will reduce the level of detail the models provide about past attacks and expand protections for users it predicts are under 18.
The first Community Safety Report is scheduled for early 2027 and will cover the second half of 2026. OpenAI plans to publish the report every six months, including anonymized examples of violence-related harms, descriptions of mitigations and policy revisions, and data assessing the effectiveness of its safeguards. The firm framed the report as a step toward greater transparency about how its models surface and respond to harmful content, and said it will complement ongoing collaboration with mental-health and behavioral experts.
Beyond publication of the report, OpenAI described concrete policy changes designed to limit the potential for its models to amplify violent behavior. The company will tighten how ChatGPT answers queries about past mass attacks and the individuals who carried them out. In practice, that means the models will provide less detailed reconstructions of incidents and will avoid granular specifics about weapons, equipment and methods for obtaining them. OpenAI also intends to restrict detailed comparisons of scale that could bolster an attacker’s notoriety. The company reiterated that its models are trained not to provide instructions that would help someone plan or execute violence.
OpenAI is also expanding the way ChatGPT responds when earlier conversations show warning signs. The models already use short factual summaries of prior warning signals to guide later responses; the company said it will apply more conservative limits when recent chats indicate possible violent intent. If a conversation shows fixation on violence or serious harm to others, ChatGPT can automatically end the chat to interrupt harmful patterns. Repeated instances of concerning behavior may lead to a temporary account pause, and accounts that violate the company’s violent-activity policies can be banned. OpenAI said it is working with external experts to refine additional automated interventions.
A notable focus of the update is protecting teen users. OpenAI will expand restrictions on weapons-related information for accounts its systems predict belong to people under 18, even when a question appears harmless. ChatGPT for Teens already provides age-appropriate safeguards; the new measures extend prohibitions on practical information about weapons and security systems that could be misused, irrespective of apparent intent.
These changes build on the age prediction system OpenAI launched the prior year. That system analyzes multiple signals, including behavior patterns, to estimate whether a user is under or over 18. OpenAI said it has strengthened the age prediction process to continually assess new signals and transition users into the under-18 experience when appropriate. Users who are incorrectly placed in the under-18 experience can verify they are adults through Persona, OpenAI’s age verification partner, which the company said preserves privacy.
Parental controls and alerts are also part of the company’s safety toolkit. Guardians can link accounts to manage safety settings, and linked accounts include alerts that notify parents when specially trained reviewers identify serious self-harm concerns or behaviors associated with disordered eating. In July, OpenAI expanded those alerts to tell parents if a teen’s account has been deactivated for violations of its violence policies, and the company said it is working to add notifications for paused accounts when repeated discussions of violence trigger enforcement.
OpenAI provided context for its enforcement work by noting it had made more than 200 law enforcement referrals in Q3 2026, and said those referrals have helped disrupt violent plots. The company characterized its role as complementary to law enforcement, families, mental-health professionals and policymakers and said it will continue iterating on technical and practical protections.
The new safeguards and reporting will begin rolling out in the coming weeks. OpenAI said it will continue investing in monitoring, rule enforcement and collaboration with real-world support services where appropriate. The combined reporting and product changes aim to make the company’s safety practices more visible while tightening guardrails around violent content and strengthening protections for younger users.
Source: Read the original source
Leave a Reply