OpenAI’s service faults make the cloud dependency visible again

July 26, 2026

OpenAI reported elevated error rates that degraded performance across its APIs, ChatGPT and Codex services. The company also logged a separate incident in which intermittent errors stopped some users from loading or continuing ChatGPT conversations, and said it had implemented mitigation while monitoring recovery.

These two incidents are likely to matter beyond one bad day because they expose the same dependency at two layers at once: the model service and the end-user application. For Microsoft, this likely sharpens a familiar constraint rather than creating a new one – if generative AI is sold as a tool inside everyday software, outages at the model provider still surface directly in the product people use. That in turn appears to strengthen the case for spreading risk across providers and for keeping more capability local where the task allows it, because reliability has again shown up as part of the product, not just the infrastructure beneath it.

On the consumer side, a counter-movement appears to be taking practical form. Public librarians in the United States began hosting heavily attended workshops on how generative AI features work and how to turn off tools such as Apple Intelligence and Google Gemini on personal devices. This likely matters less as a distribution event than as a signal that AI use is no longer just about adoption: resistance has organised itself into practical instruction.