OpenAI tightens the screws on safety to maintain away prying eyes

OpenAI has reportedly overhauled its safety operations to guard towards company espionage. In response to the Monetary Instances, the corporate accelerated an present safety clampdown after Chinese language startup DeepSeek launched a competing mannequin in January, with OpenAI alleging that DeepSeek improperly copied its fashions utilizing “distillation” strategies.

The beefed-up safety consists of “data tenting” insurance policies that restrict workers entry to delicate algorithms and new merchandise. For instance, throughout growth of OpenAI’s o1 mannequin, solely verified staff members who had been learn into the challenge might focus on it in shared workplace areas, in accordance with the FT.

And there’s extra. OpenAI now isolates proprietary know-how in offline pc methods, implements biometric entry controls for workplace areas (it scans workers’ fingerprints), and maintains a “deny-by-default” web coverage requiring express approval for exterior connections, per the report, which additional provides that the corporate has elevated bodily safety at information facilities and expanded its cybersecurity personnel.

The modifications are stated to replicate broader considerations about international adversaries trying to steal OpenAI’s mental property, although given the continued poaching wars amid American AI firms and more and more frequent leaks of CEO Sam Altman’s feedback, OpenAI could also be trying to deal with inside safety points, too.

We’ve reached out to OpenAI for remark.