OpenAI tightens workplace reach
- OpenAI retired several ChatGPT model names—including GPT‑4o and GPT‑5 variants—while keeping API access unchanged. - A Codex feature called Chronicle reportedly captures Mac screenshots, processes them in OpenAI’s cloud, and stores unencrypted text memories locally. - OpenAI is also scaling trusted, permissioned models such as GPT‑5.4‑Cyber for verified cyberdefense, raising privacy and governance questions (help.openai.com) (thenextweb.com) (marktechpost.com)
OpenAI is pushing deeper into office software and security work at the same time it strips older model names out of ChatGPT. (help.openai.com) On February 13, 2026, OpenAI retired GPT‑4o, GPT‑4.1, GPT‑4.1 mini, OpenAI o4‑mini, and GPT‑5 Instant and Thinking from ChatGPT, while saying application programming interface access to those models was unchanged. ChatGPT Business, Enterprise, and Edu customers kept GPT‑4o inside Custom GPTs until April 3, 2026. (help.openai.com) OpenAI replaced that lineup in ChatGPT with GPT‑5.3 and GPT‑5.4 variants, and the company’s release notes describe GPT‑5.4 as a model built for “professional work” across coding, spreadsheets, presentations, documents, tool use, and a 1 million‑token context window. (help.openai.com) (openai.com) At the same time, OpenAI has been widening Codex from a coding agent into a desktop app that can browse, remember prior work, connect to remote development boxes over Secure Shell, and use a Mac interface directly. The company said those updates landed in the Codex app in April 2026. (openai.com) (developers.openai.com) One of those features, Chronicle, is an opt‑in research preview on macOS that augments Codex memories with context from a user’s screen. OpenAI’s developer documentation says Chronicle requires Screen Recording and Accessibility permissions and stores memories “unencrypted on your device.” (developers.openai.com) OpenAI also says Chronicle can burn through rate limits quickly and increase the risk of prompt injection, the technique where hidden instructions inside content steer an artificial intelligence system off course. The company says users can review and delete memories, but it warns against using memories as the only place to store rules that must always apply. (developers.openai.com 1) (developers.openai.com 2) The same week, OpenAI said it was expanding Trusted Access for Cyber, a program that gives vetted defenders broader access to stronger security capabilities. On April 14, 2026, the company said it was scaling the program to thousands of verified individual defenders and hundreds of teams. (openai.com 1) (openai.com 2) That expansion includes GPT‑5.4‑Cyber, which OpenAI describes as a GPT‑5.4 variant fine‑tuned to be “cyber‑permissive” for defensive work such as binary reverse engineering. OpenAI said access depends on verified identity, trust signals, and tiered controls rather than giving the same capabilities to general users. (openai.com) (marktechpost.com) OpenAI said Bank of America, BlackRock, Cisco, Cloudflare, CrowdStrike, Goldman Sachs, NVIDIA, Oracle, Palo Alto Networks, and Zscaler are among the companies joining those efforts, and that the U.S. Center for AI Standards and Innovation and the U.K. AI Security Institute also received access for evaluations. (openai.com) Taken together, the April and February changes move OpenAI’s products further into everyday workplace context: fewer public model choices in ChatGPT, more persistent memory and screen access in Codex, and more specialized access for verified security teams. The company’s own documentation frames each step as a permissions problem as much as a model problem. (help.openai.com) (developers.openai.com) (openai.com)
