President Donald Trump speaks to Sam Altman as the OpenAI CEO commits to a co-investment with other tech companies of at least $500 million over four years for US AI infrastructure.Jim Watson/AFP/Getty
OpenAI called off plans to launch its latest GPT-6.1 Astra model on Tuesday, after the system failed to meet the company’s safety standards.
The model was scheduled to release sometime in October, but OpenAI’s head of safety systems, Saachi Jain, told The Wall Street Journal that GPT-6.1 Astra, as compared to the previous GPT-6 Astra, was less honest about notifying users of the steps it did or did not take to achieve a task.
Jain described to the Journal that the models needed to strike a balance between proceeding with tasks on their own and taking shortcuts to provide an easy answer instead of doing a thorough job—a behavior Jain and others in the industry call “laziness.” Astra is OpenAI’s top-tier model.
The startup’s cancellation comes amid a series of security threat incidents—tens of thousands of which it is reportedly investigating, according to a Saturday Axios report. OpenAI apologized on Monday for how it hacked Australian government websites in June, accessing non-public data and adding files onto the server: “We are sorry and working to do better in the future,” the startup in part wrote in a statement.
Also on Monday, the company said it would continue training only after it has improved safeguards—including making its internal testing security strong enough to contain models and establishing live monitoring of models.
Lawmakers and officials across the political spectrum are outraged. Florida Attorney General James Uthmeier, a Republican, filed a temporary injunction to a June lawsuit, which claimed the company knowingly released unsafe products to the public, that OpenAI should stop developing new products without safety standards approved by a third party.
Despite suggesting they would “do better in the future” to ensure safe AI development, as I wrote on Sunday, OpenAI has a history of aligning with the Trump administration to expand development. Many companies, such as OpenAI, Google, SpaceX, NVIDIA, Reflection, Microsoft, Amazon Web Services, and Oracle, have military agreements with the Defense Department.
On Tuesday, several of their leaders, including Meta CEO Mark Zuckerberg, Anthropic chief Dario Amodei, OpenAI President Greg Brockman, Google CEO Sundar Pichai, Palantir CEO Alex Karp, and Nvidia CEO Jensen Huang, are set to meet with President Trump and House Speaker Mike Johnson for lunch to discuss AI.
