OpenAI's enterprise controls

OpenAI is tightening how business customers get its newest artificial intelligence tools, with slower GPT-5 rollout and more features left off by default. (help.openai.com) In OpenAI’s ChatGPT Enterprise and Edu documentation, GPT-5.3 Instant and GPT-5.4 Thinking are “by default disabled” for enterprise workspaces, and admins or owners must turn them on in workspace settings. The same page says the consumer-facing retirements took effect on February 13, 2026. (help.openai.com) OpenAI’s ChatGPT Business release notes show the same pattern for connectors, which OpenAI now calls apps: updated Box, Notion, Linear, and Dropbox apps began rolling out on March 27, 2026, and admins were told to review actions and scopes before enabling them. On April 8, 2026, OpenAI added shared Outlook mailbox and calendar actions and again told workspace owners and admins to review controls before turning on the new actions. (help.openai.com) The model changes sit alongside a cleanup of ChatGPT’s public lineup. OpenAI 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, while API access remained unchanged. (help.openai.com) For business, enterprise, and education customers, OpenAI kept one exception during the transition: GPT-4o stayed available inside Custom GPTs until April 3, 2026, before full retirement across plans. OpenAI repeated that exception in multiple help articles covering model limits, GPT management, and Custom GPT creation. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) At the same time, OpenAI is widening access to a separate security-only track. On April 14, 2026, the company said it was expanding Trusted Access for Cyber to “thousands of verified individual defenders” and “hundreds of teams” and introducing GPT-5.4-Cyber, a GPT-5.4 variant tuned for defensive cybersecurity use. (openai.com) Two days later, OpenAI named early participants including Bank of America, BlackRock, Cisco, Cloudflare, CrowdStrike, Goldman Sachs, JPMorgan Chase, NVIDIA, Oracle, Palo Alto Networks, and Zscaler. It also said the U.S. Center for AI Standards and Innovation and the U.K. AI Security Institute received access to evaluate the model’s cyber capabilities and safeguards. (openai.com) OpenAI says it has been building this cyber program since 2023 through its Cybersecurity Grant Program and later added cyber-specific safeguards in 2025. The company says it has committed $10 million in API credits for the broader defense effort, including grants to groups such as Socket, Semgrep, Trail of Bits, and Calif. (openai.com 1) (openai.com 2) The result is a split rollout: fewer public model choices inside ChatGPT, more admin gates for workplace tools, and narrower access to the most sensitive cyber systems. OpenAI’s latest help pages and security posts show the company pushing new capabilities forward, but through workspace settings, verification checks, and specialized programs rather than a single open release. (help.openai.com 1) (help.openai.com 2) (openai.com)