ChatGPT Users Beware: Messages reporting Threats to Others May Be Shared with Police, OpenAI Reveals
SAN FRANCISCO – OpenAI, the creator of the popular ChatGPT chatbot, has revealed that conversations flagged as potential threats to others are reviewed by human moderators and, in serious cases, can be reported to law enforcement. The disclosure,detailed in a recent company blog post,underscores the limitations of AI safety measures and highlights that user privacy isn’t absolute when potential real-world harm is detected.
While ChatGPT is designed to offer supportive responses to users expressing suicidal thoughts, directing them to resources like the 988 hotline in the US and Samaritans in the UK, OpenAI maintains a distinct approach to threats against others. according to the company, such instances trigger a specialized review process. Trained moderators examine the chat, and if an “imminent threat” is identified, OpenAI may contact authorities. Accounts involved in making threats can also be permanently banned.
OpenAI acknowledged vulnerabilities in its safety systems, noting that safeguards are more effective in shorter exchanges and can weaken over extended conversations. The company stated it is working to improve consistency and prevent gaps in protection across multiple interactions.
Beyond threats, OpenAI is also developing interventions for other risky behaviors, including extreme sleep deprivation and hazardous stunts, aiming to ground users in reality and connect them with professional help. future plans include parental controls for teen users and options to link users with trusted contacts or licensed therapists.
The company emphasized that users should be aware that conversations are not entirely private, and messages indicating potential danger to others will be subject to review by trained moderators and could lead to police intervention.