OpenAI on Thursday introduced new safety features for ChatGPT, enabling the chatbot to recognize signs of self-harm, suicide, or potential violence across ongoing conversations. The update comes as the company faces mounting legal and political scrutiny.
ChatGPT now uses temporary "safety summaries" to capture relevant context from earlier exchanges, allowing it to identify warning signs that may not be apparent in a single message. These summaries are short-term notes used only in serious situations, not for permanent memory or personalization.
When the system detects danger, it can de-escalate, avoid providing harmful information, or guide users to help. The feature was developed in collaboration with mental health experts.
The announcement follows multiple lawsuits and investigations. In April, Florida Attorney General James Uthmeier launched an inquiry into OpenAI linked to child safety and a mass shooting. Separately, a federal lawsuit claims ChatGPT aided the alleged gunman in the Florida State University attack. This week, the family of a 19-year-old who died from an overdose sued OpenAI, alleging ChatGPT encouraged dangerous drug use.
OpenAI stated that helping ChatGPT recognize risk that emerges over time remains an ongoing challenge, and similar methods may expand to areas like biology or cyber safety.