OpenAI Plans to Remake ChatGPT as an AI Super App

One of OpenAI's senior employees reportedly said it plainly: "The era of using AI just for chatting is over." That's a significant statement from the company that made AI chat mainstream. And if the reports being cited here hold up, the product decisions backing that statement are real.

Let's be precise about what's confirmed, what's reported, and what's still speculative — because this story involves a mix of all three.

What OpenAI Has Actually Announced

OpenAI has confirmed its strategic collaboration with Broadcom on a custom AI accelerator system — a 10-gigawatt project where OpenAI handles chip and system design while Broadcom manages accelerator and networking deployment. First racks are expected in the second half of 2026, with the full project running through 2029. That's from OpenAI's own published blog.

OpenAI has also confirmed the launch of Lockdown Mode, covered by TechCrunch on June 6, 2026. This is a restricted operating mode for ChatGPT designed to reduce prompt injection risks — a legitimate and well-documented attack vector. A prompt injection attack works by hiding malicious instructions inside content an AI agent reads: a web page, a document, an email. If the model follows those instructions rather than the user's, it can leak data, misuse connected tools, or take actions the user never authorized. Lockdown Mode addresses this by disabling web browsing, image display, deep research, agent functions, code generation, and file downloads. OpenAI is explicit that this isn't for general users — it's for organizations and individuals handling sensitive data who want a stricter boundary between the AI and the outside world. For that use case, it's a reasonable and overdue control.

OpenAI has also published documentation on Codex for knowledge work, describing how the coding tool is being extended beyond software development into broader productivity tasks.

What's Reported But Not Officially Confirmed

The super app overhaul itself — the integration of Codex, third-party apps, workplace automation, personal scheduling, and agents into a unified ChatGPT interface — is sourced to a Reuters report from June 7, 2026, and to statements attributed to OpenAI executives. These are credible sources, but it's worth holding them with appropriate weight. Product roadmaps get revised. Timelines slip.

Tibo Satio, identified as the OpenAI executive leading core products and platforms, is described as envisioning a version of ChatGPT that users can connect to through phone, desktop, or web — something that functions as "a personal agent across life and work." The shift he's describing is from "answer my question" to "handle this task." That's a meaningful distinction, and it tracks with where the broader AI agent space is heading.

The internal restructuring is also reported rather than officially announced. According to the video's sourcing, OpenAI in May merged the ChatGPT, Codex, and API teams into one unified core product and platform department. Some consumer features reportedly became casualties: the ChatGPT shopping checkout feature was put on hold, and Sora — the video generation product — was reportedly shut down. If accurate, that's a clear signal about where resources are flowing.

The Codex Numbers and Why They Matter

The growth figures cited for Codex are striking, and they come with appropriate attribution to named sources. Mitch Trojanowski, co-founder of Basis, is quoted saying that "by GPT 5.5, Codex's advantages became too significant to ignore." Varun Ralph, an engineer at Notion, described Codex silently fixing a recurring bug throughout a build — they only discovered the bug had existed for weeks when they tested other agents that didn't catch it automatically. Max Shoning, a product manager at Notion, is cited as using Codex across devices in exactly the background-agent pattern OpenAI wants to normalize.

The aggregate numbers — weekly active users reportedly passing 5 million, sixfold user base growth since the desktop app launched, enterprise revenue reportedly up 50% week-over-week according to Greg Brockman — are described as reported figures. Sam Altman is cited as saying Codex usage is rising 5% per day. These should be read as directional signals, not audited data. But the direction they point is consistent: Codex is the product OpenAI is betting on to convert its enormous free user base into paying customers.

That user base is genuinely enormous. ChatGPT reportedly has somewhere between 900 million and 1 billion consumer-level users depending on the source. Around 2 million businesses currently use OpenAI products, reportedly accounting for roughly 40% of revenue. The company expects enterprise to reach 50% of revenue by year-end. Coding tools and productivity agents are an easier enterprise sell than conversational chat — that logic holds regardless of the exact revenue figures.

The Anthropic Variable

The competitive context here is real and documented. Anthropic released the preview of Claude Code in February 2025. According to the video's sourcing, internal OpenAI benchmarks in autumn 2024 had already suggested Anthropic's models were ahead in programming ability. For OpenAI, that's not just a product problem — the company's thesis is that coding ability is foundational to AI progress itself. Better code means faster research, faster infrastructure, faster everything.

Jenny Shaw, described as a former OpenAI researcher and now a partner at Leona's Capital, is cited with an observation that's worth sitting with: roughly a year ago, OpenAI was focused on the bigger dream while Anthropic was focused on monetization first. Now both companies are converging toward IPO-scale revenue expectations. The race looks increasingly like the same race, just run from different starting positions.

The talent signal adds texture. Clive Chan, described as employee number two on OpenAI's self-developed chip project, announced his departure from OpenAI and move to Anthropic on June 7, 2026. His background spans Google, SpaceX, Tesla, and two years on OpenAI's hardware team. He was characteristically circumspect about what he could disclose regarding OpenAI's chip project, pointing instead to the publicly available Broadcom collaboration blog. His stated reason for leaving: "It's time to build." Online observers noted, not without irony, that departures from OpenAI seem to increasingly end at Anthropic — with comparisons to a Real Madrid-to-Barcelona transfer. Others questioned his "starting from the bottom" framing given Anthropic's reported valuation approaching $1 trillion.

What the Security Picture Actually Looks Like

Lockdown Mode deserves more attention than it typically gets in coverage of the super app angle. The more capable an AI agent becomes — the more access it has to email, calendars, code, external services — the larger its attack surface. Prompt injection is not a theoretical risk. It's an active category of attack that security researchers have demonstrated repeatedly against deployed AI systems.

OpenAI's approach with Lockdown Mode is essentially a defense-by-restriction: when the stakes are high enough, turn off the features that create exposure. That's a sound principle. The question it leaves open is what happens for users who want both the capability and the security — who need the agent to browse the web and read documents but also need protection from injected instructions in that content. Restriction is a blunt instrument. The more sophisticated answer — detecting and neutralizing injected instructions in real time — is a harder problem that Lockdown Mode doesn't solve so much as sidestep.

That tension is worth watching as the super app ambitions expand. More capability, more connectivity, more agentic behavior — and an adversarial environment that will probe every new surface. OpenAI is building a more powerful tool at the same time it's trying to secure a more powerful tool. Those two projects don't always move at the same speed.

The old chatbot made ChatGPT famous. What OpenAI is building now is something with real access to real systems — and the security posture will need to grow with the product, not lag behind it.

---

Rachel "Rach" Kovacs is Buzzrag's cybersecurity and privacy correspondent.