I’m looking out at Alcatraz Island from a Mediterranean restaurant in San Francisco with hundred-dollar fish entrées on the menu. As I make small talk with other reporters, OpenAI CEO Sam Altman jumps through the door on my left. Altman’s looking down at his bare iPhone to show us all something, and an intrusive thought slips out of my mouth: “No phone case is a bold choice.”

Of course, I immediately realize that the billionaire CEO of OpenAI, who employs Apple veteran Jony Ive, cares more about preserving the iPhone’s original design than the $1,000 it costs to replace one.

“Listen, we’re going to ship a device that is going to be so beautiful,” says Altman, referring to OpenAI and Ive’s forthcoming AI device. “If you put a case over it, I will personally hunt you down,” he jokes.

Altman has gathered roughly a dozen tech reporters to join him and other OpenAI executives for an on-the-record dinner (and off-the-record dessert). The night raises more questions than it answers.

For instance, why is Nick Turley, the VP of ChatGPT, kindly passing me a lamb skewer just a week after launching GPT-5? Is this to encourage me to write nice things about OpenAI’s biggest AI model launch yet, which was relatively disappointing given the years of hype around it?

Unlike GPT-4, which far outpaced rivals and challenged expectations of what AI can do, GPT-5 performs roughly on par with models from Google and Anthropic. OpenAI even brought back GPT-4o and ChatGPT’s model picker, after several users expressed concerns over GPT-5’s tone and its model router.

But throughout the night, it becomes clear to me that this dinner is about OpenAI’s future beyond GPT-5. OpenAI’s executives give the impression that AI model launches are less important than they were when GPT-4 launched in 2023. After all, OpenAI is a very different company now, focused on upending legacy players in search, consumer hardware, and enterprise software.

OpenAI shares some new details about those efforts.

Altman says OpenAI’s incoming CEO of applications, Fidji Simo, will oversee multiple consumer apps outside of ChatGPT — ones OpenAI has yet to launch. Simo is slated to start work at OpenAI in just a few weeks, and she might end up overseeing the launch of an AI-powered browser that OpenAI is reportedly developing to compete with Chrome.

Altman suggests OpenAI would even consider buying Chrome — likely an offer that would be taken more seriously than Perplexity’s bid — should it become available. “If Chrome is really going to sell, we should take a look at it,” he says before looking at all of us and asking: “Is it actually going to sell? I assumed it wasn’t gonna happen.”

Simo also might end up running an AI-powered social media app — something the OpenAI CEO has said he’s interested in exploring. In fact, Altman says there’s “nothing” inspiring to him about the way AI is used on social media today, adding that he’s interested in “whether or not it is possible to build a much cooler kind of social experience with AI.”

While Turley and Brad Lightcap, OpenAI’s COO, largely give the floor to Altman, drinking wine alongside the other seated guests, Altman also confirms reports that OpenAI plans to back a brain-computer interface startup, Merge Labs, to compete with Elon Musk’s Neuralink. (“We have not done that deal yet; I would like us to.”)

How intertwined that company will be with OpenAI’s models and devices remains to be seen. Altman describes it only as a “a company that we’d invest in.”

For all the talk of browsers and brain chips, though, the elephant in the room remains GPT-5’s rough reception. Eventually, the conversation circles back to the model that has prompted our group dinner in the first place.

Turley and Altman say they’ve learned a lot from the experience.

“I legitimately just thought we screwed that up,” says Altman on deprecating GPT-4o without telling users. Altman says OpenAI will give users a more clear “transition period” when deprecating AI models in the future.

Turley also says OpenAI is already rolling out a new update to make GPT-5’s responses “warmer,” but not sycophantic, such that it won’t reinforce negative behaviors in users.

“GPT-5 was just very to the point. I like that. I use the robot personality — I’m German, you know, whatever,” says Turley. “But many people do not, and they really like the fact that ChatGPT would actually check in with you.”

It’s a delicate balance for OpenAI to strike, especially given that some users have developed dependencies on ChatGPT. Altman says OpenAI believes that less than 1% of ChatGPT users have unhealthy relationships with the chatbot — which could still be tens of millions of people.

Turley says OpenAI has worked with mental health experts to develop a rubric to evaluate GPT-5’s answers, ensuring that the AI model will push back on unhealthy behaviors.

That said, it seems that GPT-5 hasn’t hurt OpenAI’s business. In fact, Altman says OpenAI’s API traffic doubled within 48 hours of GPT-5’s launch, and the company is effectively “out of GPUs” thanks to all the demand. Cursor and other AI coding assistants have since made GPT-5 their default AI models.

In many ways, the night’s contradictions — disappointing launches, record-breaking usage — reflect OpenAI’s strange reality right now.

Given OpenAI’s bets — and others the company is making around data centers, robotics, and energy — Altman clearly has ambitions of running a much bigger company than just the ChatGPT maker. The final form could look something like Google’s parent Alphabet, but perhaps even broader.

As the night winds down, it becomes clear we aren’t gathered to reflect on GPT-5 at all. We are being pitched on a company that’s eager to outgrow its famous and controversial product.

It seems likely that OpenAI will go public to meet its massive capital demands as part of that picture. In preparation, I think Altman wants to hone his relationship with the media. But he also wants OpenAI to get to a place where it’s no longer defined by its best AI model.