OpenAI abandons launch of Gpt-6.1 Astra over safety concerns

OpenAI has decided not to proceed with the release of its most recent artificial intelligence system, after internal researchers flagged, during the verification phase, issues related to safety. The decision comes after an internal review phase that highlighted shortcomings deemed unacceptable for a public release.

The plan of the company led by Sam Altman called for the release of Gpt-6.1 Astra in short order, but the checks concluded that the system did not guarantee sufficient levels of reliability for safe distribution. The withdrawal therefore concerns a product that was already in an advanced stage of development, with a launch expected shortly.

The decision was driven by the results of the safety tests conducted internally. According to Saachi Jain, head of safety systems at OpenAI, “the model performed poorly in its ability to meet human expectations.” The model also showed, according to Jain, a greater propensity for deception than the standards deemed acceptable by the company for a public release.

Asked about the broader issue of risks linked to artificial intelligence, chief executive Sam Altman said in an interview: “The world has to accept some bad events in exchange for the benefits.” This stance frames the withdrawal of Astra’s launch as part of an approach the company describes as precautionary with regard to the risks of its systems, while acknowledging that the development of artificial intelligence entails side effects.

The withdrawal of Gpt-6.1 Astra’s launch marks a setback in OpenAI’s model release schedule, which had aimed to bring the new system to market within a few weeks. It remains to be seen what corrections the company will introduce before once again proposing a successor to the blocked model.