OpenAI tightens enterprise controls

OpenAI has started making enterprise access more restrictive, with apps disabled by default in ChatGPT Enterprise and Edu unless an admin turns them on. (help.openai.com) OpenAI’s help center says Enterprise and Edu workspace owners can enable specific apps from Workspace settings and control access with role-based permissions. A separate admin guide says those workspaces can assign app access to custom roles, while Business workspaces still have apps enabled by default. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) The change reaches beyond simple file lookups. OpenAI renamed connectors to “apps” on December 17, 2025, and those apps can search third-party services, take actions on a user’s behalf, and sync information into ChatGPT’s knowledge base. (help.openai.com) At the same time, OpenAI has been pruning the model list inside ChatGPT. Its Enterprise and Edu model guide says GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking were retired from ChatGPT on February 13, 2026, with GPT-4o in Custom GPTs fully retired across plans after April 3, 2026. (help.openai.com) The replacement has not arrived all at once. OpenAI’s release notes say GPT-5 was rolling out gradually to Plus, Pro, Team, and Free users worldwide, while Enterprise and Edu access was slated to follow; newer help pages now describe GPT-5.3 as the default for logged-in users and GPT-5.4 Thinking as the top reasoning option in ChatGPT. (help.openai.com 1) (help.openai.com 2) OpenAI has also kept some of its newest enterprise models behind admin switches. The Enterprise and Edu model page says GPT-5.3 Instant and GPT-5.4 Thinking are disabled by default for those workspaces, and admins or owners must enable them in workspace settings. (help.openai.com) That pattern now extends to cybersecurity. On April 14, 2026, OpenAI said it was scaling its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams, and introducing GPT-5.4-Cyber, a variant fine-tuned for defensive cybersecurity work. (openai.com) OpenAI said the program is aimed at vetted users responsible for defending critical software, not broad public release. The company also said firms including Cisco, Cloudflare, CrowdStrike, Goldman Sachs, NVIDIA, Oracle, Palo Alto Networks, and Zscaler had joined its cyber defense effort, alongside grants from a $10 million pool of API credits. (openai.com 1) (openai.com 2) The company’s own safety documents point in the same direction. OpenAI’s GPT-5.4 Thinking system card says the model is its first general-purpose system with mitigations for “High” cybersecurity capability, and the deployment notes say Trusted Access exists because useful defensive work can otherwise be caught by coarse account-level enforcement. (openai.com) (deploymentsafety.openai.com) Put together, the updates show OpenAI moving from broad feature launches toward tiered access: admins decide which workplace tools connect to company data, older general models are retired on a schedule, and the most sensitive cyber capabilities are reserved for verified defenders. (help.openai.com) (help.openai.com) (openai.com)