What was the Cloudbleed Bug and What Data Leaked?

Cloudbleed leaked memory across sites. Learn the cause and timeline from the official report.

Fact-checked Reviewed by Aswin Vijayan Updated November 4, 2026 Based on 1 sources
Summary

Cloudbleed was a memory leak in an HTML parser at Cloudflare. Leaks began as early as 22 September 2016 and peaked from 13 to 18 February 2017. Report data shows no malicious use and SSL private keys stayed safe. Update passwords for sites that you use and add 2-Step Verification.

Cloudbleed let a request to one site return memory from another site. People still ask what leaked and whether they must act.

Every fact below comes from the Cloudflare incident report. See the incident report on the memory leak caused by the parser bug.

What was Cloudbleed?

Cloudflare operates a network that sits in front of websites. One buffer overrun in its HTML parser sometimes returned memory that held data from other requests. Google's Project Zero reported the problem and details reached Cloudflare on 18 February 2017 at 00:32 UTC.

What Caused the Leak?

Parser code came from a Ragel source file. Cloudflare states the flaw was not in Ragel. Its error handler lacked a step and the end of buffer check used equality instead of greater than or equal. Cloudflare states the bug stayed latent for years until a newer parser changed how buffers passed between NGINX filters.

Three features used the parser chain: Email Obfuscation and Automatic HTTPS Rewrites and Server Side Excludes. Email Obfuscation was the primary cause.

What is the Timeline?

Date and Time (UTC)Event
22 Sep 2016Earliest possible leak after Automatic HTTP Rewrites was enabled
13 to 18 Feb 2017Greatest impact
18 Feb 2017, 01:19Email Obfuscation disabled
18 Feb 2017, 04:24Automatic HTTPS Rewrites disabled
18 Feb 2017, 07:22Kill switch set for Server Side Excludes
20 Feb 2017, 21:59Permanent fix went global
21 Feb 2017, 18:03Three features enabled again

What Data Leaked?

At peak about 1 in 3,300,000 HTTP requests could have leaked memory. That equals about 0.00003 percent. Report data shows no malicious exploitation.

  • Customer data. HTTP headers and POST body fragments and API JSON and URI parameters and cookies and authentication tokens like API keys and OAuth tokens.
  • Cloudflare secrets. One internal machine to machine key and a small number of internal secrets.
  • SSL private keys. These stayed safe because they live in a separate heap.
  • Search engine caches. Some leaked memory was cached under 770 unique URIs on 161 domains and Cloudflare purged them with the search engines Google and Yahoo and Bing.

How can You Respond Now?

Update passwords for important accounts and use unique ones. Add 2-Step Verification. See what to do if Gmail is hacked and man in the cloud attack prevention and website security tips.

Frequently Asked Questions

What is Cloudbleed?

Cloudbleed is a memory leak caused by a buffer overrun in an HTML parser at Cloudflare.

When did Cloudbleed Happen?

Leaks began as early as 22 September 2016 and the impact peaked from 13 to 18 February 2017.

Were SSL Private Keys Leaked?

No. Cloudflare states they were not leaked because they live in a separate heap.

Did Attackers Exploit the Bug?

No. Report data shows no malicious exploitation.

Who Found Cloudbleed?

Google Project Zero reported it to Cloudflare in February 2017.

Key Takeaways

  • Cloudbleed leaked memory from an HTML parser buffer overrun.
  • Impact peaked from 13 to 18 February 2017.
  • No malicious use turned up in the report.
  • SSL private keys did not leak.
  • Use unique passwords and 2-Step Verification.

Sources and References

All facts come from the Cloudflare incident report.

  1. Incident report on memory leak caused by Cloudflare parser bug (Cloudflare) blog.cloudflare.com