Report: AI risk surface has shifted
June 10, 2026



AI platforms are becoming part of everyday work, not just optional productivity software. Employees use ChatGPT, Claude, Gemini, and Copilot for writing, research, code, analysis and customer support, while IT teams are beginning to connect AI tools into more structured enterprise workflows.

While that growth is the opportunity, it is also what makes reliability more important. As AI systems move from short chat sessions into longer-running agentic tasks, a failed prompt, login loop, stalled code task, unavailable file, or broken connector can interrupt work that now sits inside real business processes.


Ookla analysed 471 days of US Downdetector data from January 1st 2025 through April 16th 2026 across ChatGPT, Claude, Gemini, Microsoft Copilot, AWS and Microsoft Azure, covering 3.72 million user-reported problem reports. In this research, a high-signal disruption day means a day when one service recorded more than 10 times its own median daily report volume across the period.

Key Takeaways:

AI app disruption stepped up sharply in Q1 2026. Across ChatGPT, Claude, Gemini, and Copilot, high-signal disruption days rose from six in Q1 2025 and 16 in Q4 2025 to 51 in Q1 2026. Claude accounted for 39 of those 51 service-days, while Gemini accounted for seven, Copilot three, and ChatGPT two.
OpenAI’s ChatGPT produced the largest individual AI app disruption signals, but its baseline trend has improved substantially. ChatGPT accounted for four of the five largest AI app days in the data, including roughly 68,000 reports on December 2nd 2025. Yet its daily median report volume was lower in April 2026 than in April 2025, pointing to improving reliability over time even as Codex usage has scaled rapidly in recent months.
Claude became the clearest example of scale-up volatility. Claude recorded near-zero Downdetector report volumes in early 2025, then moved into a sustained report baseline from mid-July as adoption rose. By Q1 2026, it accounted for 39 high-signal AI app disruption days, with March report volume nearly three times February’s level alone.
Hyperscaler incidents are part of the AI reliability surface. Cloud infrastructure like AWS and Microsoft Azure can create a massive blast radius when problems arise, explaining part of the AI risk surface. AWS’s October 20th 2025 DynamoDB event and Azure’s October 29, 2025 Front Door event were infrastructure-layer shocks that showed how failures in cloud control planes can quickly become visible to AI users.