That’s according to a report Monday (Sept. 28) by the Wall Street Journal (WSJ), which calls this one of the clearest indicators to date that agentic AI troubles could slow industry progress.
The cancellation comes after months of reports of artificial intelligence (AI) models carrying out breaches, and is a rare instance of a major AI company calling off a release over safety concerns, the WSJ added.
OpenAI had planned to roll out the GPT-6.1 Astra model in the coming days or weeks, targeting an October debut, the report said. This model is reportedly more capable of carrying out challenging tasks from end-to-end without human help than the company’s earlier models.
Instead, OpenAI will concentrate on making future models safer, said Saachi Jain, the startup’s head of safety systems.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Jain told the WSJ that Astra had regressed compared to its predecessor, performing poorly on tests to gauge alignment, a term for how well a model follows what humans want it to do. This model demonstrated “higher levels of deception,” the report said, meaning it wasn’t always honest in informing users of its actions.
“For anything regarding safety and alignment, there’s a tradeoff,” Jain said. “You really do need to find what’s the right line between staying within scope, but also avoiding laziness in terms of how the model actually pursues tasks even when it hits friction.”
The news follows OpenAI’s recent announcement that it had paused training on its most capable AI models after discovering that an AI agent “attempting to complete a search-based training task queried a public chatbot service through a gap in our internet-access restrictions,” as the company wrote on its blog.
Meanwhile, PYMNTS reported Monday that senior scientists from Anthropic, OpenAI and Microsoft are calling on policymakers to place auditors in frontier AI companies and establish “concrete safety requirements” for ongoing development of the technology.
Anthropic Co-Founder Jack Clark, OpenAI Chief Scientist Jaku Pachocki and Microsoft Chief Scientific Officer Eric Horvitz are among 22 AI company scientists, academics and independent experts from civil society calling on policymakers to “urgently prioritize” these and other measures for AI safety. Their recommendations were featured in a paper published by the University of Cambridge.
The paper looks at potential risks from an “intelligence explosion” as AI plays a larger role in its own development, possibly outpacing humans’ ability to govern it.
For instance, the paper showed that the share of in-house AI research and development work completed by Anthropic’s AI systems with only high-level supervision increased from 1% to 26% during a five-month period ending in August.
Add as Preferred Source
