Elevated errors for multiple models – Resolved
Multiple AI model errors elevated, affecting all users. Issue resolved.
since
Here's why. Incidents from Tuesday, September 22, 2026 (UTC), ranked byBreakage Velocity.What is Breakage Velocity?A live score of how fast an outage or regression is spreading. The more people hit it, and the faster, the higher it climbs; it cools off as things calm down.Open-ended scale: under 25 low · 25+ elevated · 100+ high · 300+ extreme.
8 outages, 3 still active today. Worst: Anthropic (<1 min).
Multiple AI model errors elevated, affecting all users. Issue resolved.
since
A hackernews post reports that enabling Chrome reading mode on an article from the Substack subdomain results in unspaced text. The original page source shows obfuscated HTML entities like <span>Will</span><span></span><span>crypto</span><span></span><span>crash</span><span></span>. This issue affects all users who use reading mode on Substack articles.
since
Error rates for ChatGPT Work across Plus, Pro, Business, Enterprise and Education plans have been elevated. The issue has now resolved.
– UTC
A disruption in Supabase connections affecting services in the EU West 1 region was resolved after a node restart. No further disruptions have been observed.
– UTC
An elevated error rate was observed for Plus and Pro users of OpenAI services, affecting Conversations. The issue has been resolved.
– UTC
Project creation operations are experiencing degraded performance in multiple regions, affecting all users. The issue has been resolved and the system is back to normal.
– UTC
Elevated error rates have been observed for requests to OpenAI's Claude Mythos 5.1 model, affecting multiple users and services. The issue is being addressed by the team.
– UTC
OpenAI reported six new incidents of concerning user behavior, which are currently under investigation. No specific technical details or affected versions were provided.
since