2025-07-12T10:15:52Z
Share
Fb
X
Copy hyperlink
lighning bolt icon
An icon within the form of a lightning bolt.
Affect Hyperlink
Save
Saved
Learn in app
This story is accessible completely to Enterprise Insider
subscribers. Develop into an Insider
and begin studying now.
Have an account? Log in.
Sam Altman mentioned OpenAI is delaying the launch of its open-weight mannequin to run extra security exams.
The open-weight mannequin, initially set for launch subsequent week, was postponed indefinitely.
The announcement comes as Grok, created by its rival xAI, grapples with its personal issues of safety.
Builders desperate to strive ChatGPT’s newest open-weight mannequin must wait — once more.Sam Altman, the CEO of OpenAI, wrote on X Friday that his firm would delay the launch to run further security exams.Initially set to debut subsequent week, the open-weight mannequin is in limbo indefinitely, Altman mentioned.we deliberate to launch our open-weight mannequin subsequent week.we’re delaying it; we want time to run extra security exams and evaluation high-risk areas. we aren’t but positive how lengthy it can take us.whereas we belief the neighborhood will construct nice issues with this mannequin, as soon as weights are…— Sam Altman (@sama) July 12, 2025
In March, Altman first mentioned an open-weight mannequin would arrive “within the coming months.” The goal slipped from June to “later this summer season” in mid-June.we’re going to take a little bit extra time with our open-weights mannequin, i.e. anticipate it later this summer season however not june.our analysis crew did one thing surprising and fairly wonderful and we predict it is going to be very very well worth the wait, however wants a bit longer.— Sam Altman (@sama) June 10, 2025
Friday’s announcement marks at the least the second delay in 4 weeks.”As soon as weights are out, they can not be pulled again,” Altman added.Open weights differ from totally open-source code. By receiving the neural community’s discovered parameters, builders can fine-tune or deploy the mannequin regionally, however not the complete coaching stack. That makes the discharge extra versatile than a closed Utility Programming Interface, but nonetheless more durable to patch as soon as copies are shared. The hold-up comes as rival xAI grapples with its personal security disaster. Enterprise Insider reported this week that xAI’s Grok chatbot posted antisemitic rants on X, prompting inner backlash and a brief posting ban.OpenAI didn’t point out Grok, and there’s no proof the 2 occasions are linked, however the episode highlights the reputational danger of releasing highly effective fashions with out strong guardrails.Elon Musk’s Grok 4 was launched on Wednesday.Builders say an open-weight GPT-4-class system from OpenAI might reshape an ecosystem now dominated by Meta’s Llama sequence and a wave of Chinese language entrants. For now, they must preserve ready whereas OpenAI runs what Altman referred to as “the brand new, extra-careful course of” to confirm its mannequin meets the corporate’s security bar.
