OpenAI Is Now Scanning Conversations and Reporting Harmful Content to Police

OpenAI has announced that it is now scanning user conversations for specific categories of harmful content, with some cases being reported to police when deemed necessary, according to OpenAI. When the system detects dialogue suggesting potential harm to others, those conversations are routed to a dedicated internal review team that can take action against accounts, including immediate suspension.

In cases where human reviewers determine an imminent threat of serious physical harm, the information may be escalated directly to law enforcement. The company emphasised that ChatGPT must not be used to inflict harm on oneself or others and that certain prohibited uses include promoting suicide or self-harm, developing weapons, or causing injury or property damage.

OpenAI also noted that self-harm disclosures are not currently shared with authorities to respect user privacy, and acknowledged that safeguards can weaken during extended conversations, with ongoing work to improve reliability and ensure responsible crisis guidance.