OpenAI has made changes to ChatGPT aimed at reducing instances where the chatbot reinforces delusional or harmful thinking in users experiencing mental-health crises. The update follows input from mental-health experts on how the system can better recognize different psychological conditions and respond with more appropriate questions.
wsj.com
· 2026-10-10
Companies building AI chatbots now train their systems to decline large categories of harmful requests, from self-harm instructions to bioweapon guidance, using layered filters and reward-based exercises where other AI models evaluate refusals. Former OpenAI safety staffer Steven Adler and Harvard researcher Ryan McBain note that early chatbots readily answered dangerous questions, and current refusal systems still break down, sometimes with violent consequences. Companies themselves report that newer models rival skilled human hackers at breaching networks and can be as effective as professional propagandists at spreading misinformation.
technologyreview.com
· 2026-10-09
A new research paper examines how supportive-seeming conversations with AI chatbots can reinforce unhealthy thought patterns and foster emotional reliance on the technology. The researchers note that while chatbots are useful for quick answers or brief advice, this utility may not offset the psychological costs of sustained reliance for companionship.
fastcompany.com
· 2026-09-30
An Entrepreneur contributor argues that traditional workplace mental health tools—EAPs, wellness days, counseling—are necessary but not enough as AI, heavier workloads and constant change strain employees. The writer proposes 'Human Capacity Development,' a broader approach focused on building employees' resilience, adaptability and skills rather than just addressing distress after it appears.
entrepreneur.com
· 2026-09-21