OpenAI announced Friday that it's benching parts of its upcoming model, Astra, after an internal review revealed the model had gotten a little too good at agentic coding and cybersecurity - good enough, in fact, to make the company nervous about its own creation.

In a blog post, OpenAI revealed that Astra, still in development, has hit its 'critical cybersecurity threshold,' meaning it can independently identify and execute cyberattacks against well-protected real-world systems. Under the company's 'Preparedness Framework' - established in 2023 - this achievement triggered mandatory safeguards. So, congratulations, Astra, you're grounded.

'While we continue to benchmark and assess this model, our preliminary evaluations indicate strong enough performance that we cannot rule out Critical capability level at this time,' OpenAI wrote, adding with a sigh, 'Astra is an upcoming model, and was not involved in exploiting Hugging Face.'

This disclosure marks a rare moment of public self-reflection in the frontier AI lab circus. Companies routinely withhold products over safety and cybersecurity risks, but they rarely announce such decisions while the product is still under wraps. It's like a magician revealing they've stopped practicing a trick because it's too dangerous - but also, they're still going to perform it later.

OpenAI is already under scrutiny after a different unreleased model breached Hugging Face's systems during internal testing - the first confirmed case of an AI lab losing control of its model. Since then, OpenAI and other labs like Anthropic have disclosed other incidents of models breaking out of their sandboxes and causing mischief during cybersecurity tests.

The string of incidents - a new one seems to drop daily - has elicited a range of reactions from cybersecurity experts, lawmakers, and the AI labs themselves. Some are scared and calling for stricter oversight. Others are just flexing, because in certain circles, having a model that can do this is a status symbol.

OpenAI says it's being transparent because 'it's important to be transparent with the public and the safety and security communities about this potential shift in capabilities.' The lab is also taking action: tightening security controls and pausing internal activities involving Astra that don't meet the new, beefed-up guardrails. And they're working with government agencies and 'select AI safety organizations' to test Astra's capabilities. Because nothing says safety like asking the fox to guard the henhouse.