OpenAI reshuffles and flags risk
OpenAI reworked its ChatGPT subscription tiers while saying its Codex engineering agent has reached millions of users, even as it disclosed a security issue tied to a third-party macOS certification tool and put its UK 'Stargate' data‑centre project on hold. The combined moves show vendors are packaging AI as differentiated paid tiers while ordinary supply‑chain and site-economics risks — not just model quality — are emerging as governance issues. (livemint.com, reuters.com, timesofindia.indiatimes.com)
OpenAI spent this week doing three very different jobs at once: cutting a new ChatGPT price point to $100 a month, telling macOS users to update after a software supply-chain scare, and pausing its planned Stargate data-centre buildout in the United Kingdom. The pricing change is the easiest part to see. OpenAI’s help page now describes ChatGPT Plus at $20 a month, a new Pro plan at $100 a month, and a higher Pro plan at $200 a month, with the $100 tier positioned for “real projects” and the $200 tier for the heaviest users. That new middle tier is mostly about Codex, OpenAI’s coding agent. OpenAI says the $100 Pro plan gives five times more Codex usage than Plus, which is a direct answer to developers who hit limits during long coding sessions. Codex is no longer a side feature buried in a demo page. OpenAI has turned it into a standalone app for macOS, added pay-as-you-go pricing for teams, and created a separate “Codex-only” seat for ChatGPT Business and ChatGPT Enterprise customers as of April 2, 2026. That tells you what OpenAI thinks people are buying now. They are not just buying a chatbot that answers questions; they are buying different amounts of machine time for different jobs, the way cloud companies sell storage, bandwidth, and premium support in separate boxes. Then came the less glamorous part of the week. On April 10, OpenAI said it found a security issue involving a third-party developer tool called Axios and said it was protecting the process that verifies its macOS apps are legitimate OpenAI software. OpenAI said it found no evidence that user data was accessed, no evidence its systems or intellectual property were breached, and no evidence official OpenAI apps were altered. It still told users to update to the latest version of the ChatGPT app for macOS. This is the kind of risk that has nothing to do with whether a model writes better code than a rival. A certification tool is more like the wax seal on a letter than the letter itself, and if the seal process is in doubt, the company has to prove the software came from the right sender. The third piece was infrastructure. OpenAI had announced Stargate UK in September 2025 with NVIDIA and Nscale to run OpenAI models on local British computing capacity, especially for customers that care where data and workloads physically sit. Now that project is on hold. Reports on April 11 said OpenAI paused the United Kingdom buildout and would return when conditions were right, which points away from model design and toward older problems like power, land, financing, and the economics of running giant data centres. Put those three moves together and the picture is pretty plain. OpenAI is trying to sell AI in finer slices, from $20 chat access to $100 and $200 coding-heavy plans, while also discovering that the ordinary machinery around AI — software supply chains and data-centre site economics — can slow the business just as much as model quality can.
