CVE-2026-47155 Vulnerability Analysis & Exploit Details

CVE-2026-47155 Vulnerability Summary

CVE-2026-47155: vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, vLLM's revision pinning controls do not consistently apply to all artifacts loaded for a model. A deployment that supplies --revision or --code-revision can still load dynamic code, GGUF files, image processors, retrieval side weights, or same-repository subfolder weights/config from an unpinned/default revision. This is a supply-chain integrity issue for pinned vLLM deployments. Operators can believe they are serving a reviewed model revision while vLLM resolves behavior-affecting nested or sibling artifacts outside that reviewed revision. This vulnerability is fixed in 0.22.0.

CVE-2026-47155 References

External References

• https://github.com/vllm-project/vllm/commit/d26a28ab033697f55a1414b5b0435de7cd6045b6
• https://github.com/vllm-project/vllm/pull/42616
• https://github.com/vllm-project/vllm/security/advisories/GHSA-3ww4-5jv9-j5gm
• https://huntr.com/bounties/3f1e24c0-87d2-4f6c-a705-820f380879ac

CWE Common Weakness Enumeration

CAPEC Common Attack Pattern Enumeration and Classification

• JSON Hijacking (aka JavaScript Hijacking) CAPEC-111 An attacker targets a system that uses JavaScript Object Notation (JSON) as a transport mechanism between the client and the server (common in Web 2.0 systems using AJAX) to steal possibly confidential information transmitted from the server back to the client inside the JSON object by taking advantage of the loophole in the browser's Same Origin Policy that does not prohibit JavaScript from one website to be included and executed in the context of another website.
• Cache Poisoning CAPEC-141 An attacker exploits the functionality of cache technologies to cause specific data to be cached that aids the attackers' objectives. This describes any attack whereby an attacker places incorrect or harmful material in cache. The targeted cache can be an application's cache (e.g. a web browser cache) or a public cache (e.g. a DNS or ARP cache). Until the cache is refreshed, most applications or clients will treat the corrupted cache value as valid. This can lead to a wide range of exploits including redirecting web browsers towards sites that install malware and repeatedly incorrect calculations based on the incorrect value.
• DNS Cache Poisoning CAPEC-142 A domain name server translates a domain name (such as www.example.com) into an IP address that Internet hosts use to contact Internet resources. An adversary modifies a public DNS cache to cause certain names to resolve to incorrect addresses that the adversary specifies. The result is that client applications that rely upon the targeted cache for domain name resolution will be directed not to the actual address of the specified domain name but to some other address. Adversaries can use this to herd clients to sites that install malware on the victim's computer or to masquerade as part of a Pharming attack.
• Content Spoofing CAPEC-148 An adversary modifies content to make it contain something other than what the original content producer intended while keeping the apparent source of the content unchanged. The term content spoofing is most often used to describe modification of web pages hosted by a target to display the adversary's content instead of the owner's content. However, any content can be spoofed, including the content of email messages, file transfers, or the content of other network communication protocols. Content can be modified at the source (e.g. modifying the source file for a web page) or in transit (e.g. intercepting and modifying a message between the sender and recipient). Usually, the adversary will attempt to hide the fact that the content has been modified, but in some cases, such as with web site defacement, this is not necessary. Content Spoofing can lead to malware exposure, financial fraud (if the content governs financial transactions), privacy violations, and other unwanted outcomes.
• Spoofing of UDDI/ebXML Messages CAPEC-218 An attacker spoofs a UDDI, ebXML, or similar message in order to impersonate a service provider in an e-business transaction. UDDI, ebXML, and similar standards are used to identify businesses in e-business transactions. Among other things, they identify a particular participant, WSDL information for SOAP transactions, and supported communication protocols, including security protocols. By spoofing one of these messages an attacker could impersonate a legitimate business in a transaction or could manipulate the protocols used between a client and business. This could result in disclosure of sensitive information, loss of message integrity, or even financial fraud.
• Application API Message Manipulation via Man-in-the-Middle CAPEC-384 An attacker manipulates either egress or ingress data from a client within an application framework in order to change the content of messages. Performing this attack can allow the attacker to gain unauthorized privileges within the application, or conduct attacks such as phishing, deceptive strategies to spread malware, or traditional web-application attacks. The techniques require use of specialized software that allow the attacker to perform adversary-in-the-middle (CAPEC-94) communications between the web browser and the remote system. Despite the use of AiTH software, the attack is actually directed at the server, as the client is one node in a series of content brokers that pass information along to the application framework. Additionally, it is not true "Adversary-in-the-Middle" attack at the network layer, but an application-layer attack the root cause of which is the master applications trust in the integrity of code supplied by the client.
• Transaction or Event Tampering via Application API Manipulation CAPEC-385 An attacker hosts or joins an event or transaction within an application framework in order to change the content of messages or items that are being exchanged. Performing this attack allows the attacker to manipulate content in such a way as to produce messages or content that look authentic but may contain deceptive links, substitute one item or another, spoof an existing item and conduct a false exchange, or otherwise change the amounts or identity of what is being exchanged. The techniques require use of specialized software that allow the attacker to man-in-the-middle communications between the web browser and the remote system in order to change the content of various application elements. Often, items exchanged in game can be monetized via sales for coin, virtual dollars, etc. The purpose of the attack is for the attack to scam the victim by trapping the data packets involved the exchange and altering the integrity of the transfer process.
• Application API Navigation Remapping CAPEC-386 An attacker manipulates either egress or ingress data from a client within an application framework in order to change the destination and/or content of links/buttons displayed to a user within API messages. Performing this attack allows the attacker to manipulate content in such a way as to produce messages or content that looks authentic but contains links/buttons that point to an attacker controlled destination. Some applications make navigation remapping more difficult to detect because the actual HREF values of images, profile elements, and links/buttons are masked. One example would be to place an image in a user's photo gallery that when clicked upon redirected the user to an off-site location. Also, traditional web vulnerabilities (such as CSRF) can be constructed with remapped buttons or links. In some cases navigation remapping can be used for Phishing attacks or even means to artificially boost the page view, user site reputation, or click-fraud.
• Navigation Remapping To Propagate Malicious Content CAPEC-387 An adversary manipulates either egress or ingress data from a client within an application framework in order to change the content of messages and thereby circumvent the expected application logic.
• Application API Button Hijacking CAPEC-388 An attacker manipulates either egress or ingress data from a client within an application framework in order to change the destination and/or content of buttons displayed to a user within API messages. Performing this attack allows the attacker to manipulate content in such a way as to produce messages or content that looks authentic but contains buttons that point to an attacker controlled destination.
• Exploitation of Thunderbolt Protection Flaws CAPEC-665 An adversary leverages a firmware weakness within the Thunderbolt protocol, on a computing device to manipulate Thunderbolt controller firmware in order to exploit vulnerabilities in the implementation of authorization and verification schemes within Thunderbolt protection mechanisms. Upon gaining physical access to a target device, the adversary conducts high-level firmware manipulation of the victim Thunderbolt controller SPI (Serial Peripheral Interface) flash, through the use of a SPI Programing device and an external Thunderbolt device, typically as the target device is booting up. If successful, this allows the adversary to modify memory, subvert authentication mechanisms, spoof identities and content, and extract data and memory from the target device. Currently 7 major vulnerabilities exist within Thunderbolt protocol with 9 attack vectors as noted in the Execution Flow.
• Browser in the Middle (BiTM) CAPEC-701 An adversary exploits the inherent functionalities of a web browser, in order to establish an unnoticed remote desktop connection in the victim's browser to the adversary's system. The adversary must deploy a web client with a remote desktop session that the victim can access.

Other 5 Recently Published CVEs Vulnerabilities

• CVE-2026-11833 – Overview: A vulnerability has been found in FAST/TOOLS and CI Server. The web server may return a response containing the CI Server setting inform...
• CVE-2026-10658 – A missing length validation in the Zephyr Bluetooth Host ISO receive path can be triggered by malformed HCI ISO data. In bt_iso_recv() (subsys/blue...
• CVE-2026-10651 – A malformed Bluetooth Classic SDP attribute can trigger a reachable assertion in Zephyr's SDP parser. In subsys/bluetooth/host/classic/sdp.c, bt_sd...
• CVE-2026-10645 – Zephyr's ext2 directory-entry parser does not fully validate on-disk directory entry structure before copying the entry name and advancing traversa...
• CVE-2026-54236 – vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a saniti...