Skip to main content

Security policy

mod_pagespeed processes untrusted content — HTML, CSS, JavaScript, and images from origin servers. We take this seriously.

Reporting vulnerabilities

If you've found a security vulnerability, please email security@modpagespeed.com with details, or use GitHub private vulnerability reporting on the module or the optimizer worker repository. Do not open a public GitHub issue for security vulnerabilities.

We do not publish a PGP key for reports. GitHub private vulnerability reporting is the encrypted channel; if you need another one, ask by email and we will arrange it before you send details.

If we emailed you about an end-of-life release

We write, once, to an organization whose public website answers with an X-Mod-Pagespeed or X-Page-Speed response header identifying a PageSpeed release that no longer receives security fixes. A notice from us looks like this:

  • • By email it comes from otto@modpagespeed.com, DKIM-signed for modpagespeed.com, as plain text without attachments, to an address published for security reports, normally your organization's own.
  • • It names the host and the header, so you can check the claim yourself, and it lists the requests we made to check it.
  • • It is sent once. It makes no offer and asks for no payment or bounty; one sentence states that we earn our income from support and consulting. We do not write again unless you reply.
  • • To confirm a notice, forward it to security@modpagespeed.com and we will tell you whether we sent it.

Before writing we request the homepage (once with the documented ?PageSpeed=off parameter), one static file linked from it, and security.txt, following redirects, and the homepage once more just before sending. We request the homepage again after 30 days and after 90 days, to see whether the header has changed. These requests carry the User-Agent weamp-notice-check/1 (+https://modpagespeed.com/security/). If a site refuses our checker, we look at its home page once in an ordinary browser, and later follow-ups use a public web crawl archive rather than a request from us. The notice lists every request we made, including any others to public pages. We do not log in, scan or test for vulnerabilities, and we do not publish host names.

If you would rather not hear from us, reply and say so: we stop the follow-up requests too and keep your domain on a do-not-contact list. How we handle the correspondence is in the privacy policy. To check a server and see what to do about it, read the end-of-life advisory.

What to include

  • • A description of the vulnerability and its potential impact.
  • • Steps to reproduce the issue.
  • • Affected versions, if known.
  • • Your assessment of severity and impact.

Our commitment

We acknowledge reports and follow a coordinated disclosure process: we work on a fix with the reporter, and after the fix is released the reporter may publish details. Critical vulnerabilities affecting production deployments jump the queue.

We keep reporters informed as we work on a fix. If a vulnerability is being exploited, we do not wait for a scheduled release: the fix is released as soon as it is ready.

Supported versions

Security fixes are made in the latest release. Upgrading to it is the supported path from any earlier release. Security fixes reach every user at the same time, through the same public channels: nobody receives a fixed build before the public release.

The original project's releases, 1.14.36.1 and earlier, receive no security fixes, and we do not maintain them. The end-of-life advisory shows how to check a server and what to do.

Safe harbor

We will not pursue legal action against security researchers who act in good faith. Good faith means: you report the vulnerability to us before disclosing publicly, you don't access or modify other users' data, and you don't degrade the service for others.

Scope

This security policy covers the mod_pagespeed software (the module, the optimizer worker, and the cache layer) on Apache, nginx and IIS, and the modpagespeed.com website.

Security updates

Every security fix is listed on the release notes, with the release that carried it: the impact, the affected and fixed versions, and whether to update, plus the CVE id when one has been assigned. The announcements describe the impact; they do not describe how to trigger an issue.

Evidence

Security disclosure

Report vulnerabilities to security@modpagespeed.com. Reports go by email or through GitHub private vulnerability reporting; the repositories' SECURITY.md files state the same policy. See security.txt.

CVE history

Tracked in the release notes.

Dependency audits

The dependency gate blocks a change on any fixable vulnerability of medium severity or higher, checked against a pinned vulnerability database, and a daily scan against the live feed catches newly published ones.

Continuous AI security scanning

mod_pagespeed has been enrolled in Anthropic’s OSS Scanner since October 2026. Its enrolment config and threat model are public.

Continuous fuzzing

mod_pagespeed has been enrolled in OSS-Fuzz, Google’s fuzzing service for open-source projects, since November 2026. Its HTML parser, JavaScript minifier, CSS parser and rule-pack code are fuzzed continuously; findings are disclosed under OSS-Fuzz’s standard policy, and reports reach security@modpagespeed.com.

SBOM

An SPDX 2.3 SBOM, with a VEX document, is published in the sbom/ directory of the source tree. The apt and yum repositories are GPG-signed and every release asset is listed in SHA256SUMS. Every release from 2.6.0 on also carries its SPDX SBOM and VEX, for everyone. How to verify the signed packages is on the support page.

Architecture note

Both parts process untrusted content by design — they rewrite HTML, minify CSS and JavaScript, and transcode images from origins you configure. We take this attack surface seriously. Both parts of mod_pagespeed build on a current toolchain and run with AddressSanitizer and UndefinedBehaviorSanitizer in CI, catching memory safety issues and undefined behavior before they reach production. The optimizer worker is written in C++23 with smart pointers and RAII throughout.

Threat model

mod_pagespeed processes untrusted images, CSS, JS, and HTML from origin servers. The module talks to the worker over a local Unix socket. The worker opens no outbound connection on its own: its network traffic is the origin fetches you configure and any key directories you enable, and its management API is off by default. All content parsing uses memory-safe patterns validated by continuous sanitizer testing (ASan, UBSan, TSan). No user data is transmitted externally. The full threat model is published in the mod_pagespeed repository as threat_model.md.

Supply chain

All dependencies are pinned with SHA-256 checksums in the build configuration.

Search