What happened during our AWS outage

Cole Gottdank· May 18, 2026

From Thursday, May 14 until midday Monday, May 18, Helicone was degraded. Requests were not being processed and the dashboard returned errors for most of that window. That is about four days, and we are not going to pretend it was shorter. This is what happened and what it means for your data.

The proxy stayed up the whole time

The most important part first: the Helicone proxy was fully functional for the entire outage. If your app sends traffic through our gateway, those requests kept flowing to your providers and your users were not affected. This was an observability outage, not a traffic outage. Your production apps did not go down because of this.

What broke

AWS flagged one of our Amazon Bedrock API keys as possibly compromised and locked our entire AWS account. The lock did not stop at that one key. It blocked our ECS tasks from starting at all, and those tasks run log ingestion and the dashboard backend. The proxy runs separately and kept serving. What broke was everything behind it: log processing, the dashboard, and auth for the dashboard. On your side that showed up as failed dashboard logins, a dashboard that would not load, and logs sitting unprocessed.

Your data is safe

We did not lose any data. Requests that arrived during the outage were queued, not dropped, and they are being processed now. Because the backlog covers roughly four days, it will take time to fully drain. Your historical data will keep filling in over the coming hours rather than all at once.

Why it lasted four days

The slow part was not the fix, it was the wait. We opened a support case the moment the account was locked and spent most of those four days waiting on AWS to respond and lift the block. We could not bring our infrastructure back up until they did, and that timing was out of our hands. We escalated repeatedly. It still took until Monday.

The key was never compromised

Once we could investigate, we confirmed the Bedrock key AWS flagged was not compromised. No breach, no unauthorized usage. We rotated every credential anyway. When a provider raises that kind of flag you assume the worst and rotate regardless of what you find, so that is what we did.

Where things stand

The account is unblocked and our services are healthy again as of Monday, May 18. No data was lost. The four-day backlog is draining and will catch up over the next while. Every key and secret has been rotated as a precaution, even though we found no sign of compromise.

This was a long outage and the wait on AWS made it worse than it should have been. Sorry. If your data still looks incomplete after the backlog has had time to drain, email us at support@helicone.ai and we will check your org directly.