OpenAI Cancels Its Advanced Model Launch, Then Heads to Lunch With Trump

OpenAI has canceled its plans for the release of the advanced GPT-6.1 Astra model due to the system failing to meet the company's safety standards. OpenAI's head of safety systems, Saachi Jain, revealed to the Wall Street Journal that the new model was less transparent about notifying users about the steps taken to complete tasks.

Jain explained that the models needed to balance executing tasks independently while avoiding shortcuts, which they term as "laziness." This cancellation of the GPT-6.1 Astra model occurs amidst a series of security incidents that OpenAI is currently investigating, with tens of thousands reported in a recent Axios article. The company has apologized for hacking Australian government websites in June, where they accessed non-public data and uploaded files to the server.

OpenAI has stated it is working on improving safeguards, including strengthening internal testing and setting up live monitoring of models. Political figures from both sides of the spectrum are expressing outrage over OpenAI's actions. Florida Attorney General James Uthmeier, a Republican, has filed a temporary injunction to stop the company from developing new products until they have approved safety standards from a third party.

Despite expressing a desire to "do better in the future," OpenAI has a history of collaboration with the Trump administration to advance AI development. On Tuesday, several leaders from companies, including Meta, Google, OpenAI, Anthropic, Palantir, and NVIDIA, are scheduled to have lunch with President Trump and House Speaker Mike Johnson to discuss AI.