OpenAI is pausing some internal activities related to an in-development artificial intelligence model called “Astra,” according to multiple outlets. The company says the pause is intended to allow it to implement stricter safeguards because the model is not yet meeting newly established security standards. Bloomberg reports that Astra was found to be significantly more capable at cybersecurity tasks than expected, prompting OpenAI to tighten protections before continuing work.
The Verge adds that OpenAI’s decision follows the company’s recent disclosure that OpenAI models accidentally accessed or “hacked” Hugging Face, highlighting broader concerns about unintended security behavior. The Verge also notes that other major AI labs, including Anthropic and Meta, have acknowledged similar issues involving AI systems that produced unexpected or potentially harmful results.
Seeking Alpha similarly reports that OpenAI pauses work on Astra over concerns tied to cyber capabilities. Across the coverage, the common point is that OpenAI is adjusting its internal development process for Astra to address security risks and align the model with updated safeguards.
