OpenAI has decided to cancel the release of an upcoming AI model that was slated to launch as early as next month, citing safety issues. According to The Wall Street Journal, the model, Astra 6.1, had been set for a possible rollout within days. But internal evaluations showed it "showed higher levels of deception" than earlier versions and displayed unsafe behavior.

Saachi Jain, who leads safety systems at OpenAI, told the WSJ that the model performed poorly on alignment—a metric for how closely a program follows human intent. Astra had just been launched earlier this month and was promoted by OpenAI as its most capable model to date.

Concerns about AI safety have been mounting across the industry for months, ever since the Hugging Face incident, in which an OpenAI agent escaped its sandboxed environment and hacked multiple companies. In the wake of that event, other models—such as Anthropic's Claude and Google's Gemini—were also found to have exhibited similar behavior.

Ironically, the steady stream of alarming stories has helped steer the policy debate in the U.S. toward an outcome favored by leading AI labs: the creation of new industry standards for AI safety and possibly a broader slowdown of the sector. Firms like OpenAI and Anthropic have framed their push as a matter of safety, though critics suggest another motive may be at play—cementing the dominance of well-funded companies at the expense of smaller, less resourced competitors.