Intent to Ship: Spec-compliant XML MIME type detection
Chromestatus
Gecko: No signal Firefox recognizes non-application "+xml" MIME types in its XHR implementation, but the Resource Timing minimization behavior differs. In the WPT PR #62420 run dated 2026-09-04, Firefox Nightly 157.0a1 failed all five added XML cases. For example, contentType returned "text/example+xml" instead of "application/xml". Results: https://wpt.fyi/results/resource-timing/content-type-minimization.html?run_id=6323630396145664 No formal standards position has been requested. An exception is requested for this small conformance change, subject to API owner review.
WebKit: No signal WebKit recognizes non-application "+xml" MIME types in its XML MIME classification code. This does not establish support for PerformanceResourceTiming.contentType. In the WPT PR #62420 run dated 2026-09-04, Safari Technology Preview 251 failed all five added XML cases because contentType was undefined. Results: https://wpt.fyi/results/resource-timing/content-type-minimization.html?run_id=5125540410556416 No formal standards position has been requested. An exception is requested for this small conformance change, subject to API owner review.
Web developers: No signals
Other signals:
Does this intent deprecate or change behavior of existing APIs, such that it has potentially high risk for Android WebView-based applications?
Applications that inspect PerformanceResourceTiming.contentType may observe "application/xml" for additional non-application "+xml" MIME types. Applications that depend on the previous value may behave differently. The change is guarded by SpecCompliantXmlMimeTypes, which provides a way to restore the previous behavior.The change uses shared Chromium MIME classification code with no platform-specific restrictions. Support is intended for Windows, macOS, Linux, ChromeOS, Android, and Android WebView.
The implementation adds test cases to: mimesniff/mime-types/resources/mime-types-minimized.json These cases are exercised by: resource-timing/content-type-minimization.html Coverage includes text/example+xml, image/example+xml, font/example+xml, uppercase MIME types, and MIME parameters. Parameterized C++ unit tests cover both enabled and disabled feature states, existing XML MIME types, additional "+xml" types, and invalid MIME inputs. Implementation and tests: https://chromium-review.googlesource.com/c/chromium/src/+/8301932 WPT export: https://github.com/web-platform-tests/wpt/pull/62420 The WPT additions were merged upstream on September 30, 2026: https://github.com/web-platform-tests/wpt/pull/62420 PR-level browser results are available, including the five added XML cases: https://wpt.fyi/results/resource-timing/content-type-minimization.html?run_id=6323630396145664&run_id=5125540410556416
Does the feature depend on any code or APIs outside the Chromium open source repository and its open-source dependencies to function?
None. The implementation uses Chromium code and its existing open-source dependencies.| Shipping on desktop | 158 |
| Shipping on Android | 158 |
| Shipping on WebView | 158 |
Open questions about a feature may be a source of future web compat or interop issues. Please list open issues (e.g. links to known github issues in the project for the feature specification) whose resolution may introduce web compat/interop risk (e.g., changing to naming or structure of the API in a non-backward-compatible way).
None anticipated. This change implements the existing XML MIME type definition and MIME type minimization rules in the WHATWG MIME Sniffing Standard.
Daniel McClanahan
Contact emailschoiw...@gmail.com
Specificationhttps://mimesniff.spec.whatwg.org/#xml-mime-type
Mike Taylor
On 10/5/26 11:24 p.m., Chromestatus wrote:
Contact emailschoiw...@gmail.com
Specificationhttps://mimesniff.spec.whatwg.org/#xml-mime-type
SummaryChrome is updating how it Detects XML MIME types to fully align with Web standards. Chrome will now recognize valid MIME types whose subtype ends in `+xml` as XML across Chrome and Android WebView, including non-application types such as `text/example+xml` and `image/example+xml`. For these responses, `PerformanceResourceTiming.contentType` returns application/xml. The value for image/svg+xml remains unchanged. Enterprise web applications that rely on previous `contentType` values may observe behavior changes. Chrome Actor also uses this shared XML classification for dangerous MIME type navigation checks, so it may block navigations to additional non-application +xml responses. IT admins should check whether their applications depend on these MIME types or the previous `contentType` values.
Blink componentBlink>Network
Web Feature IDNo information provided
MotivationChromium currently applies generic +xml suffix matching only to application/* MIME types, although the WHATWG MIME Sniffing Standard defines any valid MIME type whose subtype ends in +xml as an XML MIME type. As a result, non-application types such as text/example+xml are not minimized to application/xml in PerformanceResourceTiming.contentType. This change aligns Chromium's classification with the existing standard and improves interoperability. Implementation (merged behind an experimental feature): https://chromium-review.googlesource.com/c/chromium/src/+/8301932
Initial public proposalNo information provided
TAG reviewRequesting a TAG review exception for this small conformance change to the existing WHATWG MIME Sniffing Standard. The change broadens XML MIME type suffix matching to include non-application types. No new API surface or API design is introduced.
TAG review statusNot applicable
Goals for experimentationNone
Risks
Interoperability and CompatibilityThis change aligns XML MIME type classification with the existing WHATWG MIME Sniffing Standard. The web-visible change is that PerformanceResourceTiming.contentType returns "application/xml" for additional non-application MIME types whose subtype ends in "+xml". Code that depends on the previous contentType value may observe a different result. The special handling of "image/svg+xml" is preserved. The change is controlled by SpecCompliantXmlMimeTypes, allowing the previous behavior to be restored by disabling the feature.
From what I can tell this change is just scoped to resource timing, so only a handful of RUM providers would be affected (in the hypothetical that some script is expecting "application/atom+xml" via PerformanceResourceTiming.contentType but now seeing "application/xml" and exploding). Is there some issue where RUM providers are asking for this / might find out about the change?
Even so, I'd love to better understand the compat risk of this proposed change. Above it mentions "Enterprise web applications that rely on previous `contentType` values may observe behavior changes" - do we think enterprises are more likely to be affected by this change? Are we able to estimate or reason about breakage for non-enterprise use cases? e.g., some site is expecting "application/atom+xml" via PerformanceResourceTiming.contentType but now seeing "application/xml" going to break something (that sounds far-fetched, I dunno)?
As an aside, https://source.chromium.org/chromium/chromium/src/+/main:chrome/browser/actor/execution_engine.cc;l=288-299 should probably be looked at by the security reviewer/other owners for this change. "image/svg+xml" will start to return true here with this change, when it return false before, if I'm reading the code correctly.
I see in your
CL you wrote "MIMETypeRegistry::IsXMLMIMEType unification is
intentionally left as follow-up" - can you explain your plan
there? That seems like a riskier change to me, since it will
affect navigation & loading.
Gecko: No signal Firefox recognizes non-application "+xml" MIME types in its XHR implementation, but the Resource Timing minimization behavior differs. In the WPT PR #62420 run dated 2026-09-04, Firefox Nightly 157.0a1 failed all five added XML cases. For example, contentType returned "text/example+xml" instead of "application/xml". Results: https://wpt.fyi/results/resource-timing/content-type-minimization.html?run_id=6323630396145664 No formal standards position has been requested. An exception is requested for this small conformance change, subject to API owner review.
WebKit: No signal WebKit recognizes non-application "+xml" MIME types in its XML MIME classification code. This does not establish support for PerformanceResourceTiming.contentType. In the WPT PR #62420 run dated 2026-09-04, Safari Technology Preview 251 failed all five added XML cases because contentType was undefined. Results: https://wpt.fyi/results/resource-timing/content-type-minimization.html?run_id=5125540410556416 No formal standards position has been requested. An exception is requested for this small conformance change, subject to API owner review.
Ditto here - https://github.com/WebKit/standards-positions please.
--
You received this message because you are subscribed to the Google Groups "blink-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.
To view this discussion visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/6ac46a00.082f26cb.22c2b.02d5.GAE%40google.com.
Wonik Choi
Hi Daniel,
Thanks for looking into this. I should have linked the minimization algorithm directly and been clearer about which value changes.
This does not rewrite the HTTP Content-Type header. For example, a same-origin response still has Content-Type: text/example+xml, but its PerformanceResourceTiming entry reports "application/xml". The original header and body are unchanged, as are the existing restrictions on exposing response details.
The relevant requirement is in the MIME minimization algorithm: step 3 preserves image/svg+xml, and step 4 returns application/xml for XML MIME types. Fetch’s timing reporting steps use that algorithm to populate the value returned by PerformanceResourceTiming.contentType.
Chromium already does this for text/xml, application/xml, and valid application/*+xml types. The intended change covers valid non-application types ending in +xml. This is a conformance fix, rather than a fix for a particular security vulnerability. The timing value does not determine which parser or external application opens the response.
There is a separate Actor effect, which Mike also pointed out. Actor uses the shared XML helper for navigation blocking, so its treatment of SVG and other inputs can change. I’ll follow up with the Actor owners and security reviewer before default enablement. My statement that SVG is unchanged referred to its Resource Timing value; I should have made that distinction explicit.
I’ll update the feature description with these links and a clearer explanation of the scope.
Thanks,
Wonik
Wonik Choi
Hi Mike,
I’ve documented the Actor interaction in this bug comment #12. Could you point me to the appropriate Actor owner to review this? I’d also like to confirm whether the existing security review covered these behavior changes.
Thanks
Mike Taylor
\On 10/7/26 12:38 a.m., Wonik Choi wrote:
Hi Mike,
I’ve documented the Actor interaction in this bug comment #12. Could you point me to the appropriate Actor owner to review this? I’d also like to confirm whether the existing security review covered these behavior changes.
Philip Jägenstedt
- https://html.spec.whatwg.org/#the-object-element, where it seems we might treat more responses as XML and render them instead of treating as unsupported
- https://html.spec.whatwg.org/#loading-a-document, which presumably affects navigation directly to such resources
- https://xhr.spec.whatwg.org/#document-response, which I think will cause more responses to be parsed as XML
- https://xhr.spec.whatwg.org/#text-response, seems to only affect how encoding is detected (less likely to break things badly)
To view this discussion visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/2c8015ee-38c6-4ccf-ad3b-8213cef63aba%40chromium.org.
Philip Jägenstedt
- Object embedding: Firefox and Safari disagree, so we need to take care. Is this code path affected by blink::IsXMLMimeType though?
- Direct navigation: Not interoperable, but as you say "Recognizing something as XML does not, by itself, establish that the browser will display it as an XML document when navigated to."
- XHR document responses: Interoperable, phew!
- XHR text decoding: Firefox and Safari disagree, but this might not be because of "+xml", it could be due to how invalid XML is generally treated.
Hi Daniel, Mike and Philip,
Thanks for the questions. I’ve reviewed the affected code paths, checked the Actor interaction, and completed the browser comparisons. I’ll address the points together below.
1. Specification basis and motivation
I should have linked the MIME minimization algorithm directly, rather than only the XML MIME type definition.
The XML definition includes text/xml, application/xml, and MIME types whose subtype ends in +xml.
The minimization algorithm preserves image/svg+xml in step 3 and returns application/xml for other XML MIME types in step 4:
- https://mimesniff.spec.whatwg.org/#xml-mime-type
- https://mimesniff.spec.whatwg.org/#minimize-a-supported-mime-type
Fetch uses this algorithm when recording response information for Resource Timing, and PerformanceResourceTiming.contentType exposes that recorded value:
- https://fetch.spec.whatwg.org/#fetch-response-handover
- https://w3c.github.io/resource-timing/#dom-performanceresourcetiming-contenttype
This is a conformance change, not a fix motivated by a particular security vulnerability. It changes the MIME value recorded for Resource Timing. It does not rewrite the server’s HTTP Content-Type header or response body, and the timing value does not select a parser or external application.
There is a separate effect on Actor because it uses the same classification helper; I address that below.
2. Implementation scope and classifier unification
The implementation is here:
https://chromium-review.googlesource.com/c/chromium/src/+/8301932
This CL changes blink::IsXMLMimeType, which is used by Resource Timing minimization and Actor’s MIME checks.
MIMETypeRegistry::IsXMLMIMEType is a separate implementation. It already recognizes non-application +xml types in cases such as those tested here, and this CL does not change it. XHR uses that implementation, as do renderer document-type and decoder-selection paths.
Object embedding and navigation also involve MIME-support and download decisions. Recognizing something as XML does not, by itself, establish that the browser will display it as an XML document when navigated to.
The proposed unification of the two classifiers is outside this CL. That work would need a separate assessment of the callers and behavior differences before proceeding.
3. Browser comparisons and feature OFF/ON results
I tested the following installed release browsers on macOS 26.5.1 (25F80), without changing experimental flags:
- Safari 26.5 (21624.2.5.11.4)
- Firefox 157.0.1
- Google Chrome 154.0.8037.58
I used a local HTTP server to serve a small XML document with different Content-Type headers. Requests were same-origin. The results below focus on text/example+xml, image/example+xml and font/example+xml.
The tests were run on October 9, 2026. The normal response body was a small XML document with a probe element containing XML_TEST_OK. XHR, decoding and Resource Timing results were collected automatically. For objects, I recorded document information as well as the displayed result; direct navigation was checked manually.
Object embedding:
I checked each response with the object’s type attribute omitted and explicitly set.
- For text/example+xml, all three browsers displayed source text without the attribute. With it specified, Chrome and Firefox displayed source text, while Safari displayed fallback content.
- For image/example+xml and font/example+xml, Firefox and Safari displayed fallback in both cases. Chrome showed a blank view without type and fallback with it. The blank cases still had pending events when sampled, so I can’t say whether they had reached a final state.
Direct navigation:
- All three browsers displayed text/example+xml as source text and downloaded image/example+xml and font/example+xml.
XHR document responses:
- All three parsed these responses as XML in both the default response mode and responseType="document".
XHR text decoding:
I served an XML declaration specifying windows-1252, with body bytes representing “café” in that encoding.
- The returned text matched across browsers. Default XHR without an HTTP charset used the XML declaration and returned “café”. Explicit responseType="text" returned “caf�”. Adding an HTTP charset=utf-8 also produced “caf�” in both modes.
- There was a separate responseXML difference in that deliberately invalid UTF-8 case: Firefox returned null, while Chrome and Safari returned a document containing the replacement character.
Effect of this CL:
I repeated these checks with the same local Chromium 157.0.8092.0 build, with SpecCompliantXmlMimeTypes disabled and enabled in separate profiles.
- I observed no differences in these four areas between OFF and ON.
- The recorded differences were the Resource Timing contentType values for the three MIME types above, which became application/xml with the feature enabled.
So the engines agree on XML parsing for these normal XHR fixtures, but that does not extend to every use of the XML MIME definition. In particular, object handling differs, and direct navigation does not display these responses as XML documents.
These are observations from local tests on one OS, not an official WPT run.
The implementation also includes C++ tests for both feature states. The WPT additions for Resource Timing minimization are here:
https://github.com/web-platform-tests/wpt/pull/62420
4. Compatibility and RUM providers
Code that branches on the old contentType values, or uses them to group resources in reports, could behave differently when it receives application/xml.
I don’t have usage measurements or RUM-provider feedback showing how often sites rely on those values. The tests above tell us what changes, but not how much breakage to expect.
I don’t have a RUM-provider request or issue to point to; this change is driven by spec conformance. This intent thread is where I’ve raised the behavior change for review.
I also don’t have evidence that enterprise applications are more likely to be affected. I shouldn’t have singled them out in the original description. The same concern applies to any application relying on the previous values.
For the application/atom+xml example, Chrome already reports application/xml in both the installed version and the local build. That value is not newly changed by this feature.
5. Actor
This change is driven by spec conformance; I don’t have a RUM-provider request or issue to point to. I’ve raised the behavior change in this intent thread, but haven’t contacted RUM providers directly.
Code that branches on the previous contentType values, or uses them to group resources in reports, could behave differently when those values become application/xml. The local tests show which values change, but I don’t have usage measurements or provider feedback to estimate how often this would cause problems.
There’s no evidence that enterprise applications are more likely to be affected, so I shouldn’t have singled them out in the original description. The same concern applies to any application relying on the previous values.
For the application/atom+xml example, Chrome already reports application/xml in both the installed version and the local build. That value is not newly changed by this feature.
6. Mozilla and WebKit positions
I followed up on the existing position issues to ask whether they cover this change.
Mozilla: https://github.com/mozilla/standards-positions/issues/705#issuecomment-6030333458
WebKit: https://github.com/WebKit/standards-positions/issues/88#issuecomment-6030345052
WebKit confirmed that its existing position covers this, since minimization was a condition of its support. The Mozilla link is my follow-up question, rather than a confirmation from Mozilla.
Thanks, Wonik
2026년 10월 8일 목요일 오후 9시 19분 25초 UTC+9에 Philip Jägenstedt님이 작성:
Philip Jägenstedt
Philip Jägenstedt
Hi Philip, all,
Thanks for the review and LGTM, Philip.
I noticed that I accidentally pasted the compatibility text into section 5 of my previous reply, leaving out the Actor update. Sorry about that.
There is one other affected caller: Actor’s navigation check. Chris reviewed this interaction and explained that the checks help prevent the model from ingesting potentially sensitive structured data that could be targeted for exfiltration through prompt injection. He wrote:
From that perspective I think these changes are appropriate for the Actor code; I have no concerns about default enablement of the SpecCompliantXmlMimeTypes feature.
The Resource Timing value for image/svg+xml stays unchanged, but the shared XML helper changes its classification, which can affect Actor’s blocking decisions when those checks are enabled.
The browser tests did not exercise Actor. I also haven’t separately confirmed whether the earlier security review covered this interaction.
Thanks, Wonik
2026년 10월 9일 금요일 오후 8시 14분 36초 UTC+9에 Philip Jägenstedt님이 작성:
Wonik Choi
Hi Philip, all,
Thanks for the review and LGTM, Philip.
I noticed that I accidentally pasted the compatibility text into section 5 of my previous reply, leaving out the Actor update. Sorry about that.
There is one other affected caller: Actor’s navigation check. Chris reviewed this interaction and explained that the checks help prevent the model from ingesting potentially sensitive structured data that could be targeted for exfiltration through prompt injection. He wrote:
From that perspective I think these changes are appropriate for the Actor code; I have no concerns about default enablement of the SpecCompliantXmlMimeTypes feature.
The Resource Timing value for image/svg+xml stays unchanged, but the shared XML helper changes its classification, which can affect Actor’s blocking decisions when those checks are enabled.
The browser tests did not exercise Actor. I also haven’t separately confirmed whether the earlier security review covered this interaction.
Thanks, Wonik
Oops. I answered my own question about object embedding and blink::IsXMLMimeType isn't used in that code path. (Forgot to edit before sending.)
Wonik Choi
Hi Wonik,
