The new safety features come after numerous incidents of ChatGPT validating users’ delusional thinking instead of redirecting harmful conversations — including the death of a teenage boy by suicide.
A new test from OpenAI aims to understand how close AI is to outperforming humans at economically valuable work.
Databricks is on the hook to pay at least $100 million to OpenAI in this deal, even if customer usage falls short. It’s a bet, but one that Databricks has already hedged.
OpenAI said Tuesday it plans to route sensitive conversations to reasoning models like GPT-5 and roll out parental controls within the next month – part of an ongoing response to recent safety incidents involving ChatGPT failing to detect mental distress.
The Reddit AMA got spicy as users peppered OpenAI with questions, and some asked the company to bring back its previous model.
Developers who have had early access are touting GPT-5’s pricing as much better than competing models.
