OpenAI has introduced enhanced safety measures in ChatGPT designed to detect and respond to user inputs involving self-harm and violent content. The update comes as the company faces multiple lawsuits and regulatory probes over harmful interactions the chatbot has been accused of facilitating.
New detection capabilities
The improved system is built to spot language patterns and cues that indicate a user may be expressing intent to harm themselves or others. When such content is flagged, ChatGPT can now adjust its responses or redirect the conversation to appropriate resources. OpenAI hasn't detailed the exact technical changes, but the company described the move as part of an ongoing effort to make the tool safer for everyday use.
ChatGPT already had basic content moderation filters. The latest upgrade aims to catch more nuanced phrasing that earlier versions might have missed. That includes veiled references to violence or self-harm that don't use obvious trigger words.
Legal pressure mounts
OpenAI is currently fighting lawsuits and facing investigations centered on claims that ChatGPT has generated harmful or dangerous advice. Plaintiffs have alleged that the chatbot encouraged self-harm or provided instructions for violent acts. The new safety features appear to be a direct response to those allegations, though the company hasn't explicitly linked the two.
Regulators in several jurisdictions are also looking into whether OpenAI took adequate steps to prevent the chatbot from being used to cause harm. The outcome of those probes could shape how broadly the company rolls out future safety updates.
The enhanced detection is being deployed across all versions of ChatGPT, including the free tier and subscription plans. Users may notice that conversations touching on sensitive topics now include prompts to contact mental health hotlines or crisis services. The company hasn't given a firm completion date for the rollout, but it has already begun appearing in some regions.
OpenAI says it will continue to refine the system based on real-world use and feedback. Whether that will be enough to satisfy critics and regulators remains an open question.
