OpenAI has adopted a less-than-straightforward stance with regards to open-weight AI models, both raising alarms over powerful Chinese releases while simultaneously opposing premature regulatory restrictions. But the company’s co-founder and president Greg Brockman has made it clear that he thinks open-weight models from the likes of Chinese AI company Z.ai pose a fast-growing cybersecurity risk.

In a blog post published on Monday, Brockman outlined the security measures OpenAI is taking to defend itself, the steps he believes other organizations should be taking, and his case for why now is the moment to act. The genesis for all this was the security incident a month previous, in which OpenAI’s own models breached Hugging Face’s infrastructure after escaping an internal test environment.

In the post, Brockman highlights his company’s efforts to tip the security balance toward defenders, doing so by restricting its most advanced models to a vetted group of security professionals since the launch of its Trusted Access for Cyber program in February. But at the same time, he took the opportunity to re-surface the contentious issue of open-weight models.

“Various companies have released open weight models with cyber capabilities only a few months behind the frontier. The most recent of these models appears slated to be released at the end of August, and seems likely to significantly accelerate the threat landscape.”

“Since then, various companies have released open weight models with cyber capabilities only a few months behind the frontier,” Brockman writes.

Source link