DOMPurify: Trusted Types policy survives `clearConfig()` and can poison later `RETURN_TRUSTED_TYPE` output
- When
- Where
- Global (internet)
- Category
- cyber_advisory · npm
## Impact A DOMPurify instance that is reused across trust boundaries can stay bound to a previously supplied `TRUSTED_TYPES_POLICY` even after `clearConfig()` is called. A later caller that requests `RETURN_TRUSTED_TYPE` receives a `TrustedHTML` object created by the old policy, not by a clean default configuration. If the old policy is unsafe or controlled by a less-trusted integration, this turns a later "default" sanitize call into script execution at a Trusted Types sink. `TRUSTED_TYPES_POLICY: null` on the later call also does not clear the retained policy. [dompurify-trusted-types-policy-survives-clearconfig-poc.js](https://github.com/user-attachments/files/28604913/dompurify-trusted-types-policy-survives-clearconfig-poc.js) ## Affected version Tested against DOMPurify `3.4.8`, repository commit `825e617753ac1169306a542d3174a77f717a0cf6`. ## Root cause `_parseConfig()` overwrites `trustedTypesPolicy` when `cfg.TRUSTED_TYPES_POLICY` is truthy, but the default/null path only initializes the internal policy when `trustedTypesPolicy === undefined`. Once a custom policy has been set, later default config parsing leaves it in place. Relevant code: - `src/purify.ts:786-812` accepts and stores `cfg.TRUSTED_TYPES_POLICY`. - `src/purify.ts:813-832` does not reset an existing policy when config has no policy or has `TRUSTED_TYPES_POLICY: null`. - `src/purify.ts:2123-2125` signs the final serialized HTML with the retained policy when `RETURN_TRUSTED_TYPE` is true. - `src/purify.ts:2133-2136` `clearConfig()` only clears `CONFIG` and `SET_CONFIG`; it does not reset `trustedTypesPolicy` or `emptyHTML`. ## Local PoC Run from the DOMPurify checkout, or set `DOMPURIFY_REPO`: ```bash node /home/dompurify-trusted-types-policy-survives-clearconfig-poc.js ``` Observed output: ```json { "result": { "baseline": "<b>baseline</b>", "duringPolicy": "<img src=x onerror=alert(\"TT_POLICY_SURVIVED_CLEARCONFIG\")>", "afterClearString": "<img src=\"x\">", "afterClearTrustedType": "[object TrustedHTML]", "afterClearTrusted": "<img src=x onerror=alert(\"TT_POLICY_SURVIVED_CLEARCONFIG\")>", "afterNullTrusted": "<img src=x onerror=alert(\"TT_POLICY_SURVIVED_CLEARCONFIG\")>", "mountedHTML": "<img src=\"x\" onerror=\"alert("TT_POLICY_SURVIVED_CLEARCONFIG")\">" }, "dialogs": [ "TT_POLICY_SURVIVED_CLEARCONFIG" ] } ``` The important part is the split behavior after cleanup: - `purify.clearConfig(); purify.sanitize(...);` returns a normal sanitized string (`<img src="x">`), because the later call is not asking for a Trusted Type. - `purify.clearConfig(); purify.sanitize(..., { RETURN_TRUSTED_TYPE: true });` still uses the old policy and returns attacker-controlled `TrustedHTML`. - Passing `{ TRUSTED_TYPES_POLICY: null, RETURN_TRUSTED_TYPE: true }` also still returns attacker-controlled `TrustedHTML`. ## Preconditions This is a shared-instance state contamination issue. It matters when one DOMPurify instance is reused by multiple integrations, plugins, request handlers, or components with different trust levels, and a cleanup step relies on `clearConfig()` to restore safe defaults. This is not a default string-input bypass. An attacker must be able to influence a prior `TRUSTED_TYPES_POLICY` on the reused instance, or a less-trusted integration must have installed an unsafe policy. ## Severity impact is XSS at a Trusted Types sink in applications that reuse a DOMPurify instance across trust boundaries. Attack complexity is high because exploitation depends on prior policy injection or a less-trusted integration and a later `RETURN_TRUSTED_TYPE` sink. ## Suggested fix Make `clearConfig()` reset Trusted Types state as part of restoring defaults, or have `_parseConfig()` explicitly clear `trustedTypesPolicy` and `emptyHTML` when `TRUSTED_TYPES_POLICY: null` is supplied.
Sources
- GitHub Advisory Database ↗ · first seen 2026-06-15 20:12 UTC
Defaxon links out to the original reporting and never republishes article text.
Correlated events
Computed by the Defaxon correlation engine — linked by shared actors, co-location, and temporal proximity. Scored hypotheses, never causal claims.