CVE-2026-48710
Kludex Starlette HTTP Request/Response Smuggling Vulnerability
Description
CVE-2026-48710 is a medium-severity request interpretation vulnerability in Starlette, the lightweight ASGI framework that underpins FastAPI and many Python web services. Prior to version 1.0.1, the HTTP Host request header was not validated before being used to reconstruct request.url; because Starlette's routing algorithm relies on the raw HTTP path while request.url is rebuilt from the Host header, a malformed header could make request.url.path differ from the path that was actually requested. Middleware and endpoints that enforce security restrictions based on request.url rather than the raw scope path could therefore be bypassed. Publicly tracked as the "badhost" vulnerability, CVE-2026-48710 carries a CVSS score of 6.5 but an exceptionally high EPSS score of roughly 36% at the 98th percentile, and it is listed in the CISA Known Exploited Vulnerabilities (KEV) catalog.
KEV Information
CVSS Score
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:NOpen in CalculatorAffected Products
| Vendor | Product | Version |
|---|---|---|
| encode | starlette | >= 0.8.3, < 1.0.1 |
| redhat | ai inference server | >= 3.3.0, <= 3.3.5 |
| redhat | ansible automation platform | 2.6; 2.7 |
| redhat | migration toolkit for applications | < 8.2.0 |
| redhat | openshift ai | >= 3.3, < 3.3.5; >= 3.4, < 3.4.2 |
| redhat | openshift lightspeed | - |
| redhat | satellite | 6.17; 6.18; 6.19 |
| redhat | enterprise linux ai | 3.0 |
Multiple CVSS Assessments
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N
References
- https://badhost.org/(Mitigation, Third Party Advisory)
- https://github.com/Kludex/starlette/commit/764dab0dcfb9033d75442d7a359645c9f94648c6(Patch)
- https://github.com/Kludex/starlette/security/advisories/GHSA-86qp-5c8j-p5mr(Vendor Advisory)
- https://github.com/pypa/advisory-database/tree/main/vulns/starlette/PYSEC-2026-161.yaml(Third Party Advisory)
- https://ostif.org/disclosing-the-badhost-vulnerability-in-starlette(Mitigation, Third Party Advisory)
- https://www.secwest.net/starlette(Exploit, Mitigation, Third Party Advisory)
- https://www.x41-dsec.de/lab/advisories/x41-2026-002-starlette(Exploit, Mitigation, Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:22992(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:22993(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:23346(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:24866(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:26226(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:30088(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:30089(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:34456(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:34526(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:34532(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:37275(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:43038(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:44696(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:51357(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:60520(Third Party Advisory)
- https://access.redhat.com/errata/RHSA-2026:63337(Third Party Advisory)
- https://access.redhat.com/security/cve/CVE-2026-48710(Third Party Advisory)
- https://bugzilla.redhat.com/show_bug.cgi?id=2481742(Issue Tracking, Third Party Advisory)
- https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-48710.json(Third Party Advisory)
- https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-48710(Third Party Advisory, US Government Resource)
- https://www.wiz.io/blog/ai-infrastructure-honeypot(Exploit, Third Party Advisory)
- https://www.microsoft.com/en-us/security/blog/2026/08/26/when-ai-infrastructure-becomes-target-securing-gateways-control-points(Broken Link)
Weakness Type
CWE-444: Inconsistent Interpretation of HTTP Requests ('HTTP Request Smuggling')
This weakness arises when different components interpret the boundaries or meaning of an HTTP request differently, so a request that looks harmless to one layer means something else to another. In CVE-2026-48710 the inconsistency is internal to the application stack: the router reads the raw path from the ASGI scope, while request.url is reconstructed from the attacker-influenced Host header, so a security check reading one representation and the router acting on the other can be made to disagree.
Learn more: CWE-444 — Inconsistent Interpretation of HTTP Requests ('HTTP Request Smuggling')
CWE-1289: Improper Validation of Unsafe Equivalence in Input
This weakness occurs when a product accepts an input used as a resource identifier without validating that it is not equivalent to a dangerous value in some alternative form, letting attackers slip past filters with a variant spelling. Starlette accepted a Host header that did not conform to the RFC grammar and used it to construct a URL, so the resulting request.url.path could be an alternative representation that a path-based security filter did not recognise as the same resource.
Learn more: CWE-1289 — Improper Validation of Unsafe Equivalence in Input
Impact Analysis
CVE-2026-48710 is remotely exploitable over the network with low attack complexity, no authentication and no user interaction — the attacker simply sends a request with a malformed Host header. Both confidentiality and integrity impacts are Low and availability is unaffected, because the flaw does not directly disclose data or execute code; it bypasses whatever restriction the application implemented on top of request.url. That makes the real impact entirely dependent on the deployment: an application whose authentication middleware, admin-path restriction or rate limit keys off request.url.path loses that control, while one that uses the raw scope path is unaffected — which is why the CVSS base score of 6.5 (Medium) may understate the risk in specific stacks. The striking figure is the EPSS score of approximately 36% at the 98th percentile, one of the highest currently assigned, reflecting that Starlette is embedded in an enormous number of internet-facing Python services and that the attack requires only a single crafted header; CISA's KEV listing confirms this is being exploited in the wild.
Exploit Maturity
CVE-2026-48710 has strong exploitation signals across every dimension. CISA has confirmed active exploitation through its Known Exploited Vulnerabilities listing with a 16 September 2026 deadline, and public exploit material is available from multiple independent sources: X41 D-Sec's advisory X41-2026-002 and the SecWest write-up are both tagged as exploit references, OSTIF published a disclosure of the badhost vulnerability, and a dedicated site at badhost.org documents the issue. The upstream fix commit is public, so the vulnerable reconstruction logic can be identified by diff. Most telling is the EPSS score of roughly 36%, in the 98th percentile — an extraordinary value that reflects both the trivial nature of the attack, a single malformed header, and Starlette's presence beneath a very large share of Python web services. Treat every unpatched Starlette deployment as actively targeted.
Remediation
- Follow the CISA KEV required action. Apply mitigations in accordance with the vendor's instructions, comply with CISA's BOD 26-04 "Prioritizing Security Updates Based on Risk" guidance and the "Forensics Triage Requirements", and discontinue use of the product if mitigations are unavailable. The KEV due date is 16 September 2026.
- Upgrade Starlette to 1.0.1 or later. Versions from 0.8.3 up to but excluding 1.0.1 are affected; 1.0.1 validates the
Hostheader against the grammar of RFC 9112 §3.2 and RFC 3986 §3.2.2 when constructingrequest.urland falls back toscope["server"]for malformed values. Details are in GHSA-86qp-5c8j-p5mr and the fix commit. - Patch downstream products that bundle Starlette. Red Hat AI Inference Server 3.3.0–3.3.5, Ansible Automation Platform 2.6 and 2.7, Migration Toolkit for Applications before 8.2.0, OpenShift AI 3.3.x and 3.4.x, OpenShift Lightspeed, Satellite 6.17–6.19 and Enterprise Linux AI 3.0 are listed as affected; apply the corresponding Red Hat errata (RHSA-2026:22992, RHSA-2026:22993 and RHSA-2026:23346) rather than only upgrading direct dependencies.
- Audit your own code for
request.url-based security decisions. Grep the application and its middleware for authorization, path allowlisting, redirect validation and rate limiting that readsrequest.urlorrequest.url.path, and switch those checks to the raw ASGIscope["path"], which the router itself uses — this removes the inconsistency regardless of framework version. - Normalise and validate
Hostat the edge, and harden long term by rejecting or rewriting non-conformingHostheaders in the reverse proxy or ingress in front of the application, configuring an explicit allowed-hosts list, and adopting the general defence for this class: make every layer of the request chain interpret the request identically, and validate identifier inputs with a strict allowlist rather than by blocking known-bad spellings.
Technical Details
CVE-2026-48710 is an inconsistent-interpretation flaw (CWE-444) enabled by improper validation of unsafe equivalence in input (CWE-1289) in Starlette. Starlette routes requests using the raw HTTP path carried in the ASGI scope, but exposes a convenience request.url object that it reconstructs partly from the Host request header. Before version 1.0.1 that header was consumed without validating it against the HTTP and URI grammars, so a malformed value could steer URL reconstruction such that request.url.path no longer matched the path the router had used. Any middleware or endpoint that made a security decision from request.url — blocking /admin, enforcing authentication on a prefix, applying per-path rate limits — therefore evaluated a different string than the one that determined which handler ran, and the restriction could be bypassed while the request still reached the protected route. The CVSS vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N reflects an unauthenticated, low-complexity network attack with limited direct impact, since the severity of the bypass depends on what the application protected that way. The 1.0.1 fix validates the header against RFC 9112 §3.2 and RFC 3986 §3.2.2 and falls back to scope["server"] when it is malformed.
Frequently Asked Questions
Is CVE-2026-48710 being actively exploited?
Yes. CISA added CVE-2026-48710 to its Known Exploited Vulnerabilities catalog with a remediation due date of 16 September 2026. Multiple public exploit write-ups exist — from X41 D-Sec, SecWest, OSTIF and the badhost.org disclosure site — and the EPSS score of about 36% places it in the 98th percentile, among the highest exploitation probabilities currently assigned.
What products are affected by CVE-2026-48710?
Starlette is affected from version 0.8.3 up to but excluding 1.0.1, which means FastAPI and other frameworks built on it are affected transitively. Red Hat products bundling Starlette are also listed, including AI Inference Server 3.3.0–3.3.5, Ansible Automation Platform 2.6 and 2.7, Migration Toolkit for Applications before 8.2.0, OpenShift AI 3.3.x/3.4.x, OpenShift Lightspeed, Satellite 6.17–6.19 and Enterprise Linux AI 3.0.
How do I fix CVE-2026-48710?
Upgrade Starlette to 1.0.1 or later, which validates the Host header against RFC 9112 and RFC 3986 and falls back to scope["server"] for malformed values, and apply the corresponding vendor errata for any bundled product. Also audit your middleware for security decisions based on request.url and switch them to the raw scope["path"].
How severe is CVE-2026-48710?
CVE-2026-48710 is rated Medium with a CVSS 3.1 base score of 6.5, reflecting low direct confidentiality and integrity impact and no availability impact. The practical severity depends on your application: if middleware enforces authentication or path restrictions using request.url, those controls can be bypassed entirely — and with an EPSS score in the 98th percentile and confirmed KEV-listed exploitation, it should be patched with high urgency regardless of the base score.
Need Help With Vulnerability Management?
Our security experts can help you prioritize and remediate vulnerabilities effectively.