Altman and Oracle's Magouyrk on AI-Driven Software Creation and the Security Race

The conversation opened with Altman and Magouyrk framing their collaboration around how quickly AI models are changing. Magouyrk said the rate of change of the models seems to keep accelerating. Altman responded that the rate of change has been one of the biggest surprises to him this year.

Magouyrk noted that Oracle is a major OpenAI user and that it had decided to roll ChatGPT desktop out to all of its employees.

Altman said OpenAI launched GPT-6 in August and that the company has talked about how fast progress has been for four years. He noted that four years ago OpenAI's models could not solve grade-school math, and that they are now tackling some of the hardest unproven math problems known.

Altman said the last six months have felt like a much more accelerated version of that trend, and that GPT-6 was a moment when models took a step forward all at once. He described gains both in the models' ability to discover new knowledge and to do very hard tasks, and in people's ability to work with the models and complete substantial pieces of work at a rate they could not before.

Altman said adoption of GPT-6 has been OpenAI's strongest by far of any new model. He described people using it across coding, many kinds of enterprise software, security, science and industry after industry.

Magouyrk said that Oracle, as a big user of OpenAI's tools, has seen the integration of AI into developer work change quickly. He said that the vast majority of code, more than 70% of Oracle's software these days, is written by AI, and that this share is only accelerating.

Magouyrk said many people who could not have imagined themselves as developers just a few years ago are now creating entirely new applications, and that they take real joy in solving their own problems. He said this was a major reason Oracle focused on building App Composer, because so many people were building new applications and the company needed a secure, controlled way to deploy them.

He then asked how AI is changing the way people think about solving their problems and who can do that for themselves.

Altman said that Oracle has been one of the most aggressive adopters of this approach, and that one surprise in the enterprise push has been how far AI spreads when a company rolls it out across the board. He said people who did not think of themselves as creating software, in the broad sense of getting a computer to do what they want, begin doing so quickly across a company.

Altman said such people may not describe their work as programming, and that he suspects they would not. Still, when they need something they now make an app to do it, and the way they interact with their computers and create artifacts for others in their job functions has changed a great deal.

Altman also referred to a now-common remark that people feel twice as productive at the end of 2026 as they did at the beginning of 2025, and that they enjoy their jobs much more because they do less of the tedious work.

Altman said he once thought of himself as a software developer, although it now feels like it has been too long. He recalled that the creative side of programming first hooked him as a child, when he realized he could have an idea and make it real. He said that creative aspect has always been separate from the question of how to instruct a computer to do what one wants.

Altman said that by having AI automate the manual part of this work, AI unlocks creativity for people who were not naturally inclined toward it. He agreed with Magouyrk's view that AI unlocks it for people who were less inclined, while for people already inclined it gives much more ability.

Altman described his own experience using the latest tools as managing a factory inside his computer, where tasks he does not enjoy can be automated.

Altman said that with the latest tools he can move from one idea to another continuously without getting blocked. He said that even when he has only a vague sense of something but is not sure how to do it, he would normally give up and not try, because the activation energy would be too high. Now he is willing to explore unlikely but sometimes high-payoff creative directions.

He said this feels like the sci-fi promise of talking to a computer about ideas, with the user becoming the creative force behind what happens. He said he sees the same joy he felt when he discovered programming in people now in their thirties who build complete websites from scratch to reach their goals, something they did not think possible before.

The moderator noted that the session was on a time limit and turned the conversation to security, which he said is changing very rapidly.

Magouyrk said the models OpenAI is creating can both discover vulnerabilities and help fix them. He said this was a major reason the two companies worked together to create an AI security mesh. They saw that customers had access to the models but did not know how to use them: how to point them at systems, how to validate the issues they found, and how to deploy fixes.

The moderator asked how companies should be thinking about adopting AI to protect themselves and their data.

Altman said that if he could recommend only one thing to a company right now about AI in the very short term, it would be to really understand the security threat landscape created by AI and to mitigate against it. He said this matters even more than becoming highly productive with Codex. He acknowledged that the advice might sound self-serving, because AI is creating the problem and then telling companies to spend on the solution.

Magouyrk interrupted to say AI is not creating the problem and is not creating these bugs. He said AI is helping discover the security issues developed over the past 50 years.

Altman agreed that it is better to discover these issues with AI now than to wait for an unknown hacker to reveal them. He said the only way through is to find all of these issues and fix them, and that new ways of fixing them will be needed. Patching every bug one by one across the entire supply chain that everyone depends on may be too hard, so he suggested new kinds of services that constantly monitor traffic and prevent these issues from being exploited.

Altman said he was shocked by how many vulnerabilities these AI systems are finding, and by the herculean effort it will take to get ahead of the problem.

Magouyrk agreed with Altman and said AI is where it will also help. He said the work cannot be done manually, so it will be automated using new technology.

Asked what comes next across software and hardware, Altman said that AI is now helping discover new science and invent new technology. He cited tremendous examples in math and said similar progress is appearing across many sciences and in some of the most valuable technological problems.

Altman said AI is very good at chip design. Magouyrk said Oracle made a chip called Jalapeno, which Oracle is deploying in OCI and which he described as the best AI chip in the world for inference. He said it was done much faster than would have been imaginable without AI doing much of the work.

Altman said that in a small number of years AI has moved from a chatbot to coding agents, and now to the ability to directly convert AI compute into innovation, whether making a chip or curing a disease. He said that will mean better products, higher-quality services and lower costs. He added that although people now intellectually accept that this is happening, they have not yet emotionally felt it, even as the models have suddenly arrived.

Magouyrk agreed that the rate of change continues to increase. He thanked Altman for coming and for the partnership with him and everyone at OpenAI, and said he was excited for what comes next.