Join our Newsletter — 33% off our NHI Course
Home› Guides› Permissions Policy, COOP, COEP and CORP Guide: Isolate…
Guide Web & API Security

Permissions Policy, COOP, COEP and CORP Guide: Isolate Browser Contexts Without Breaking Login

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 27 October 2026 12 min read
On this page

Content Security Policy controls what a page may load, but it does not stop another site holding a handle to your window, embedding your private resources in its own process, or using the camera API through an embedded third-party frame. Four response headers fill that gap: Cross-Origin-Opener-Policy (COOP), Cross-Origin-Embedder-Policy (COEP), Cross-Origin-Resource-Policy (CORP) and Permissions-Policy. They are among the least understood security headers, partly because the wrong value quietly breaks sign-in popups, payment flows or passkeys. This guide explains what each header does, how they combine into cross-origin isolation, which values to use for each type of response, how to roll them out with report-only modes, and how to measure them. It is for web application security engineers and front-end platform teams. For the full header set, start with the Security Headers Guide; this page goes deeper on these four.

Key takeaways

  • Each header answers a different question. COOP decides who can hold a reference to your window, CORP decides who can embed your resource, COEP decides what your page may embed, and Permissions-Policy decides which powerful browser features your page and its frames may use.
  • Most sites need COOP and CORP, not COEP. COEP is only required for cross-origin isolation, which unlocks SharedArrayBuffer and higher-precision timers; COOP and CORP protect almost every site against cross-window and cross-origin read attacks.
  • Login and payment flows need same-origin-allow-popups. Plain COOP same-origin severs the window reference that many OAuth and payment popups rely on.
  • Copying a maximal Permissions-Policy can disable passkeys. An empty allowlist for publickey-credentials-get blocks WebAuthn even on your own top-level page, and WebAuthn calls reject with NotAllowedError.
  • Report first, then enforce. COEP and Permissions Policy have report-only variants and send violations to Reporting-Endpoints, so breakage is visible before users see it.

The threat model behind the headers

The W3C Post-Spectre Web Development draft states the problem plainly: Spectre-style side channels mean active content can read any data that enters the address space of the process hosting it, and browsers cannot yet reliably isolate same-site windows or framed origins into separate processes. Its core principle is that data must not unexpectedly enter an attacker's process. A second family of attacks, cross-site leaks (XS-Leaks), uses window references, frame counts, timing and error behaviour to infer what a logged-in user sees on another site. See the XS-Leaks Guide and the Side-Channel Attacks Guide.

HeaderSet onProtects againstDefault without the header
Cross-Origin-Opener-PolicyHTML documentsCross-origin pages keeping a window.opener or window.open() handle to your document; XS-Leaks through that handleunsafe-none: share a browsing context group with anyone
Cross-Origin-Resource-PolicyAny resource (images, scripts, API responses)Other sites pulling your private resources into their process with no-cors requestsEmbeddable as though cross-origin
Cross-Origin-Embedder-PolicyHTML documents and workersYour own page loading cross-origin resources that never consented; a prerequisite for isolationunsafe-none
Permissions-PolicyHTML documents (plus iframe allow)Injected or embedded code using camera, microphone, geolocation, payment, WebAuthn and other powerful featuresEach feature's default allowlist, * or self

Cross-Origin-Opener-Policy

COOP controls whether a new top-level document opens in the same browsing context group (BCG) as its opener or in a new one. In a new group, references between the two windows are severed and the documents may be process-isolated. MDN documents four values:

  • unsafe-none: the default; the document can share its group with any other document.
  • same-origin: the document only shares a group with same-origin documents that also send same-origin. Required for cross-origin isolation.
  • same-origin-allow-popups: like same-origin, but popups the document opens with window.open() stay in the group if they send unsafe-none. MDN names cross-origin OAuth and payment services as the use case.
  • noopener-allow-popups: always opens into a new group, even from a same-origin opener, so a same-origin page cannot script it. Useful for isolating a sensitive area (for example an admin console) from the rest of the same origin.

The outcome depends on both documents. On navigation the documents stay together only if both send unsafe-none, or both send the same value and are same-origin. With window.open(), an opener on same-origin-allow-popups keeps a reference to a popup sending unsafe-none, which is exactly what popup sign-in libraries need.

OWASP ASVS 5.0 requirement V3.4.8 asks that every response that initiates document rendering sends COOP with same-origin or same-origin-allow-popups as required.

Cross-Origin-Resource-Policy

CORP is set by the resource owner and tells the browser to block no-cors cross-origin or cross-site requests for that resource. It has three values: same-origin, same-site and cross-origin. It complements CORS rather than replacing it: CORS grants other origins the ability to read a response; CORP stops them embedding it via image, script or similar tags where no CORS check happens. ASVS V3.5.8 asks that authenticated resources can only be loaded or embedded on the user's behalf when intended, using a restrictive CORP or strict Sec-Fetch validation. See the Fetch Metadata and Resource Isolation Policy Guide and the CORS Misconfiguration Guide.

Resource typeCORPCOOPNotes (from the W3C post-Spectre baselines)
Private API responses, avatars, user uploadssame-originNot applicableAdd X-Content-Type-Options: nosniff and a Vary on Sec-Fetch headers if you branch on them
Public static assets on a CDN (fonts, scripts, images)cross-originNot applicableBroad sharing is fine; CORS * and Timing-Allow-Origin: * are reasonable
Resources meant to be embedded but not read cross-origincross-origin, no CORS headersNot applicableEmbeddable, not readable
Sensitive documents (account settings)same-originsame-originPlus framing protection; COEP only if the page needs isolation-gated APIs
Sign-in and payment pages that open popupssame-originsame-origin-allow-popupsKeeps popup references working
Documents designed to be opened by other sitesAs neededunsafe-none, sent explicitlyThe draft recommends sending permissive values explicitly so future secure defaults do not break you

Cross-Origin-Embedder-Policy and cross-origin isolation

COEP sets your document's policy for loading cross-origin resources requested in no-cors mode. MDN documents three values:

  • unsafe-none: the default; anything loads.
  • require-corp: cross-origin no-cors resources load only if they send a CORP header that allows it; resources requested in cors mode (for example with the crossorigin attribute) must pass CORS instead.
  • credentialless: cross-origin no-cors resources load without a CORP header, but the request is sent without cookies and cookies in the response are ignored.

Send one token only: MDN notes that multiple tokens or multiple headers behave as unsafe-none. A document is cross-origin isolated when it is served with COOP same-origin and COEP require-corp or credentialless, and the cross-origin-isolated Permissions Policy feature (default allowlist self) is not blocked. self.crossOriginIsolated then returns true and the page can share SharedArrayBuffer through postMessage(), gets better Performance.now() precision and can call Performance.measureUserAgentSpecificMemory().

Decision guidance: choose credentialless when the page embeds many third-party resources you cannot get CORP headers on and those resources do not need cookies. Choose require-corp when you control the embedded origins or need credentialed cross-origin subresources. Do not enable COEP at all if the page does not use isolation-gated APIs: the cost is breakage and the security gain over COOP plus CORP is small.

Permissions-Policy

Permissions Policy, formerly Feature Policy, lets a site declare which browser features may be used by the page and by each embedded frame. The W3C specification (a Working Draft on the Recommendation track, dated 22 September 2026 when checked) defines the header as a Structured Field dictionary mapping feature tokens to allowlists. Unrecognised features are ignored. Allowlist values, per MDN:

AllowlistMeaningExample
*Allowed in the document and every frame, any originfullscreen=*
()Disabled everywhere, including the top-level pagecamera=(), microphone=()
selfAllowed for the document and same-origin framesgeolocation=(self)
Named originsAllowed for listed origins; wildcard subdomains where supported (https://*.example.com does not match https://example.com)payment=(self "https://pay.example")
src (iframe allow only)Allowed if the frame's document matches its src origin; the default in allow<iframe allow="camera">

Two layers decide what a frame gets: the parent's header and the iframe's allow attribute (the container policy). A feature must be allowed by both. Features not named in the header fall back to their default allowlist, either * or self, so a policy that omits a feature does not disable it. When a policy blocks a powerful feature, the user is not prompted and the Permissions API reports denied.

A sensible baseline disables features the site never uses, for example camera=(), microphone=(), geolocation=(), usb=(), serial=(), hid=(), and grants anything needed to named origins only. The OWASP Secure Headers Project publishes a much broader recommended list that sets almost every feature to (); read it as a starting point, not a drop-in value, because it also disables payment, fullscreen and publickey-credentials-get.

Where identity fits

These headers interact directly with authentication flows, which is why they are often rolled back after an incident ticket rather than tuned:

  • Popup sign-in and federated login: OAuth and OIDC popups, and some payment providers, return results to the opener window. COOP same-origin on the relying party page severs that reference; use same-origin-allow-popups there, as both MDN and the W3C draft advise. Redirect-based flows avoid the issue entirely. See the OAuth for Browser-Based Applications and the BFF Pattern Guide.
  • Passkeys and WebAuthn: publickey-credentials-get has a default allowlist of self. Setting it to () blocks navigator.credentials.get() on your own login page, which rejects with NotAllowedError. If sign-in runs in a cross-origin iframe, grant the feature to that origin in both the header and the iframe allow attribute. See the FIDO2 and WebAuthn Internals Guide.
  • Authenticated resources: CORP same-origin on session-bound API responses and user content stops other sites pulling authenticated data into their process.
  • Isolating admin areas: COOP noopener-allow-popups on privileged consoles stops a same-origin page compromised by XSS from scripting an admin window it opened.

Rolling out the headers

  1. Inventory flows that cross origins: popup logins, payment widgets, embedded video, chat widgets, analytics, documents opened by partners, and any page that is iframed.
  2. Start with CORP on private resources: set same-origin on API responses and user uploads, and cross-origin explicitly on public assets. This rarely breaks anything you own.
  3. Add COOP: same-origin on most document responses, same-origin-allow-popups on pages that open sign-in or payment popups. Test every login method, including social login and passkeys, in every supported browser.
  4. Add Permissions-Policy in report-only: send Permissions-Policy-Report-Only with a report-to endpoint and collect permissions-policy-violation reports before enforcing.
  5. Only if needed, add COEP: deploy Cross-Origin-Embedder-Policy-Report-Only with require-corp and a Reporting-Endpoints entry, fix or switch to credentialless, then enforce.
  6. Enforce in rings and keep a rollback: set headers at the edge or gateway so a single configuration change reverts them; do not hard-code them in many applications.

Small versus large estates

A small site with one origin and a handful of third parties can usually ship CORP, COOP and a short Permissions-Policy in a single release after testing its login flows. A large estate has many origins, shared CDNs, partner integrations and pages that other teams iframe, so it needs a header policy per response class (documents, private APIs, public assets, embeddable widgets), owned centrally and applied at the edge, with each application able to request a documented exception. In both cases, keep the policy in version control next to the edge configuration, so a reviewer can see what changed when a flow breaks.

Testing that the headers work

  • Window handle test: from a cross-origin test page, open your document and confirm the returned handle can no longer reach it (for example closed reads as true once COOP separates the groups).
  • Embedding test: from a cross-origin page, embed a private resource with an image or script tag and confirm the browser blocks it under CORP.
  • Feature test: call a disabled feature such as geolocation from the page and from a frame and confirm it is denied without a prompt.

Detection and measurement

  • Header coverage: percentage of document responses with an explicit COOP value, and of authenticated responses with CORP, from synthetic scans or edge logs.
  • Violation report volume: COEP and Permissions Policy reports per day, broken down by blocked URL or feature; a spike after a release means a new third party.
  • Isolation status: where you depend on isolation, a client metric recording crossOriginIsolated true or false.
  • Authentication success rate: sign-in completion for popup and passkey flows before and after each header change.
  • Unused powerful features: count of features the site never uses that are still allowed by default.

Framework mapping

FrameworkReferenceRelevance
OWASP ASVS 5.0V3.4.8COOP on responses that render documents
OWASP ASVS 5.0V3.5.8CORP or Sec-Fetch validation for authenticated resources
OWASP ASVS 5.0V3.2.1Preventing content rendering in an incorrect context
NIST SP 800-53 Rev. 5SC-39 Process IsolationBrowser process separation that COOP and COEP enable
NIST SP 800-53 Rev. 5CM-7 Least FunctionalityDisabling unused browser features with Permissions-Policy
NIST CSF 2.0PR.PS-06Secure development practice for header baselines

Common failure patterns

  • COOP same-origin on the login page: popup sign-in silently fails, the header is removed in a hurry, and never comes back.
  • Pasting a maximal Permissions-Policy: passkeys, payments or fullscreen video break because every feature is set to ().
  • Assuming omission disables a feature: unlisted features keep their default allowlist, often * or self.
  • COEP everywhere: enabling require-corp on pages that never needed isolation breaks third-party embeds for no benefit.
  • Duplicate headers: the application and the CDN both set COEP, producing multiple values that behave as unsafe-none.
  • CORP same-origin on public CDN assets: blocks your own subdomains and partners; use same-site or cross-origin for shared assets.

Practitioner checklist

  • Send COOP explicitly on every HTML response, choosing same-origin-allow-popups for pages that open sign-in or payment popups.
  • Set CORP same-origin on authenticated API responses and user content, and explicit values on public assets.
  • Write a Permissions-Policy that disables features you never use and grants the rest to named origins.
  • Confirm publickey-credentials-get and payment are allowed wherever passkeys and payments run.
  • Use report-only modes and a Reporting-Endpoints collector before enforcing.
  • Enable COEP only on pages that need cross-origin isolation, and verify crossOriginIsolated.
  • Set headers in one place (edge or gateway) and test for duplicates.
  • Add sign-in success rate to the release checks for any header change.

Standards and references

Related NHI Mgmt Group resources: Security Headers Guide · XS-Leaks Guide · Fetch Metadata and Resource Isolation Policy Guide · Content Security Policy Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 27 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.