On April 11, the Financial Times revealed that OpenAI has dramatically curtailed the time and resources allocated to safety testing for its advanced AI models, sparking fears that its technology may be prematurely deployed without adequate scrutiny. Notably, the safety assessment period for OpenAI's latest large language model has been drastically cut down to mere days, in stark contrast to previous iterations. Both internal staff and third-party teams have voiced concerns regarding the diminished thoroughness of the testing process under these changes.

According to eight insiders familiar with OpenAI's testing protocols, the $300 billion startup is facing pressure to swiftly roll out new models to sustain its competitive edge. This has resulted in a significant lack of time and resources being dedicated to risk identification and mitigation. These insiders further disclosed that OpenAI aims to launch its new model, o3, as soon as next week, affording testers less than a week for safety checks. This is a stark departure from previous testing phases that typically spanned several months.