---
title: "Elevated OpenAI API errors — June 11, 2026"
date: 2026-06-12
excerpt: On June 11 we observed elevated HTTP 431 errors from the OpenAI API across client accounts. OpenAI has since resolved the incident, but our logs show the issue beginning roughly 90 minutes before their official start time.
---

# Elevated OpenAI API errors — June 11, 2026

On June 11, 2026 Coolhand detected elevated HTTP 431 (Request Header Fields Too Large) errors from the OpenAI API across client accounts. OpenAI has since resolved the incident. Here's what we observed.

## What happened

Starting around **20:44 UTC on June 11**, Coolhand's request logs began showing 431 responses from OpenAI's API. Both error responses included a header not present in prior successful requests: `x-openai-proxy-wasm: v0.1`, pointing to a newly deployed WebAssembly proxy layer on OpenAI's infrastructure.

The errors affected clients using different codebases, different SDK versions (openai-node 4.86.2 and 5.12.2), and different request structures. The only common factor was OpenAI's infrastructure.

Importantly, the request headers were not actually large. Standard headers totaling roughly 400 bytes were being misclassified as oversized by the new proxy. This was a proxy bug, not a client-side issue.

## The timeline discrepancy

OpenAI's status page ([incident 01KTWCER83NNKE698QXNXJG11M](https://status.openai.com/incidents/01KTWCER83NNKE698QXNXJG11M)) lists the incident start as **22:22 UTC** on June 11. Our first affected log entry is timestamped **20:44 UTC** — approximately 90 minutes earlier.

This is consistent with a rolling infrastructure deployment: the WASM proxy hit some API servers before others, causing isolated failures before the issue became widespread enough to trigger an official incident declaration.

## Resolution

OpenAI fully resolved the incident at **00:47 UTC on June 12**. We have not seen 431 errors from OpenAI's API since.

## What this means for your integration

No client-side code changes are needed. If you want to guard against transient proxy errors like this in the future, adding retry logic for 431 responses is a reasonable precaution — the OpenAI SDK does not retry 4xx status codes by default.