OpenAI Community Safety Report and Enhanced Teen Protections

Overview of Safety Initiatives

OpenAI is formalizing its commitment to safety through the introduction of a biannual Community Safety Report and the implementation of more stringent content restrictions. These updates aim to address the risks of AI being used to facilitate real-world violence, building on existing systems that identified over 200 credible threats for law enforcement referral in Q3 2026.

Community Safety Report

Starting in early 2027, OpenAI will publish a Community Safety Report covering the second half of 2026, with subsequent updates every six months. This report will provide transparency into the company’s safety infrastructure, including:

  • Methodologies for training systems to detect, investigate, and respond to community safety issues.
  • Anonymized examples of how violence-related harm manifests in user interactions.
  • Data-backed insights into the effectiveness of current mitigations.
  • Documentation of new policy changes and safety interventions.

Policy Restrictions on Violent Content

OpenAI is tightening its usage policies to prevent the misuse of ChatGPT for planning or researching violent acts. These restrictions apply to all users regardless of age or intent:

  • Mass Attack Reconstructions: The model will provide less detailed information regarding past mass attacks. This includes restricting content that reconstructs the events, details specific weapons or equipment used, explains how those items were obtained, or provides comparisons of harm that could amplify an attacker's notoriety.
  • Contextual Awareness: ChatGPT is being updated to recognize warning signs across multiple turns of a conversation. If a user exhibits patterns of fixating on violence, the system will respond more conservatively and may automatically terminate the chat. Repeated violations of this nature may lead to temporary account suspension.

Enhanced Protections for Teens

Recognizing the need for age-appropriate safeguards, OpenAI is expanding restrictions for users identified as under 18 through its age prediction systems. Key updates include:

  • Weapons Information: Even when queries appear harmless, the system will now restrict practical information regarding weapons and security systems that could be misused.
  • Parental Oversight: OpenAI is expanding its parental control features. Parents with linked accounts will be notified if their teen's account is paused due to repeated discussions of violence, adding to existing alerts for serious self-harm concerns or policy violations.
  • Age Verification: Users who are incorrectly flagged by the age prediction system as being under 18 retain the ability to use the Persona verification service to confirm their adult status and remove these additional restrictions.

Sources