OpenAI reports 200 law enforcement referrals after Q3 2026

OpenAI frames this announcement as an expansion of work already underway, not a new direction. Company policy already bars models from giving instructions that help someone plan or carry out violence, and OpenAI says it relies on mental health and behavioral experts to assess difficult cases. The 200-plus law enforcement referrals in Q3 2026 are presented as evidence that this pipeline is producing results.

The Community Safeguards Report is the transparency piece. OpenAI says it will explain how its systems are trained to detect, investigate, and respond to community safety issues, share anonymized examples of violence-related harm, and include data supporting its findings. The first report, due in early 2027 for the second half of 2026, will set the baseline against which later editions are read.

Separately, OpenAI is tightening content boundaries. It will restrict responses about past mass attacks and their perpetrators, limiting reconstructions such as weapon and equipment details and comparisons of scale of harm, regardless of a user's age. Those content restrictions, plus the teen weapons limits and automated chat-ending for fixation on violence, are slated to begin rolling out within the next few weeks.