Wasmtime wasi-http EUVD-2026-115863

Exploitation requires that the host embed Wasmtime's wasi-http implementation and permit an untrusted/malicious guest to initiate outbound HTTP requests - the guest is the attacker and supplies the multiple conflicting content-length headers. Critically, smuggling only materializes when the forwarded request reaches a downstream server or intermediary that does NOT treat conflicting content-length values as an error; strict/compliant intermediaries that reject such messages neutralize the attack. This downstream dependency is the limiting factor captured by AC:H/AT:P, and it means the impact is on the HTTP path rather than the Wasmtime host itself. Affected versions are < 36.0.19, >= 37.0.0 and < 48.0.7, and >= 49.0.0 and < 49.0.4; the complete fix is updating to 36.0.19, 48.0.7, or 49.0.4 (or main/50.0.0), after which guests may set only a single content-length value.

This is a moderate, conditionally-exploitable integrity issue rather than a top-priority emergency. The supplied CVSS 4.0 vector (AV:N/AC:H/AT:P/PR:N/UI:N) scores 6.3 and models exploitation as network-reachable but gated by both high complexity (AC:H) and a present attack requirement (AT:P), with impact limited to low integrity on the vulnerable system (VI:L) and no confidentiality or availability impact (VC:N/VA:N). Those AC:H/AT:P metrics align with the mechanism: the host emits conflicting content-length headers, but actual smuggling only succeeds against a downstream server or intermediary that fails to reject the malformed message - a condition outside the attacker's direct control. The vuln.today decision label is 'this_month' (not act_now/emergency), consistent with a real-but-not-urgent priority. No CISA KEV listing and no EPSS score are present in the input, and no exploit code is referenced - only the GHSA and VulnCheck advisories - so there is no active-exploitation or weaponization signal to elevate this. Note one cross-source inconsistency: the input carries an 'Information Disclosure' tag, yet both the CVSS vector (VC:N) and CWE-444 (HTTP request smuggling) point to an integrity rather than a confidentiality problem - treat the information-disclosure framing as unverified.

A host embeds Wasmtime and lets an untrusted WebAssembly guest make outbound HTTP requests via wasi-http. The malicious guest sets two conflicting content-length headers on an outbound request; the host forwards it, and a downstream proxy or server that does not reject conflicting content-length values interprets the message boundary differently from its upstream, allowing a smuggled request. No public exploit code is referenced in the input, so exploitation is theoretical at the time of analysis and further depends on a permissive intermediary.