CVE-2026-25640 Vulnerability Analysis & Exploit Details

CVE-2026-25640 Vulnerability Summary

CVE-2026-25640: Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 to before 1.51.0, a path traversal vulnerability in the Pydantic AI web UI allows an attacker to serve arbitrary JavaScript in the context of the application by crafting a malicious URL. In affected versions, the CDN URL is constructed using a version query parameter from the request URL. This parameter is not validated, allowing path traversal sequences that cause the server to fetch and serve attacker-controlled HTML/JavaScript from an arbitrary source on the same CDN, instead of the legitimate chat UI package. If a victim clicks the link or visits it via an iframe, attacker-controlled code executes in their browser, enabling theft of chat history and other client-side data. This vulnerability only affects applications that use Agent.to_web to serve a chat interface and clai web to serve a chat interface from the CLI. These are typically run locally (on localhost), but may also be deployed on a remote server. This vulnerability is fixed in 1.51.0.

CVE-2026-25640 References

External References

• https://github.com/pydantic/pydantic-ai/releases/tag/v1.51.0
• https://github.com/pydantic/pydantic-ai/security/advisories/GHSA-wjp5-868j-wqv7

CWE Common Weakness Enumeration

CAPEC Common Attack Pattern Enumeration and Classification

• XSS Using MIME Type Mismatch CAPEC-209 An adversary creates a file with scripting content but where the specified MIME type of the file is such that scripting is not expected. The adversary tricks the victim into accessing a URL that responds with the script file. Some browsers will detect that the specified MIME type of the file does not match the actual type of its content and will automatically switch to using an interpreter for the real content type. If the browser does not invoke script filters before doing this, the adversary's script may run on the target unsanitized, possibly revealing the victim's cookies or executing arbitrary script in their browser.
• DOM-Based XSS CAPEC-588 This type of attack is a form of Cross-Site Scripting (XSS) where a malicious script is inserted into the client-side HTML being parsed by a web browser. Content served by a vulnerable web application includes script code used to manipulate the Document Object Model (DOM). This script code either does not properly validate input, or does not perform proper output encoding, thus creating an opportunity for an adversary to inject a malicious script launch a XSS attack. A key distinction between other XSS attacks and DOM-based attacks is that in other XSS attacks, the malicious script runs when the vulnerable web page is initially loaded, while a DOM-based attack executes sometime after the page loads. Another distinction of DOM-based attacks is that in some cases, the malicious script is never sent to the vulnerable web server at all. An attack like this is guaranteed to bypass any server-side filtering attempts to protect users.
• Reflected XSS CAPEC-591 This type of attack is a form of Cross-Site Scripting (XSS) where a malicious script is "reflected" off a vulnerable web application and then executed by a victim's browser. The process starts with an adversary delivering a malicious script to a victim and convincing the victim to send the script to the vulnerable web application.
• Stored XSS CAPEC-592 An adversary utilizes a form of Cross-site Scripting (XSS) where a malicious script is persistently "stored" within the data storage of a vulnerable web application as valid input.
• Cross-Site Scripting (XSS) CAPEC-63 An adversary embeds malicious scripts in content that will be served to web browsers. The goal of the attack is for the target software, the client-side browser, to execute the script with the users' privilege level. An attack of this type exploits a programs' vulnerabilities that are brought on by allowing remote hosts to execute code and scripts. Web browsers, for example, have some simple security controls in place, but if a remote attacker is allowed to execute scripts (through injecting them in to user-generated content like bulletin boards) then these controls may be bypassed. Further, these attacks are very difficult for an end user to detect.
• AJAX Footprinting CAPEC-85 This attack utilizes the frequent client-server roundtrips in Ajax conversation to scan a system. While Ajax does not open up new vulnerabilities per se, it does optimize them from an attacker point of view. A common first step for an attacker is to footprint the target environment to understand what attacks will work. Since footprinting relies on enumeration, the conversational pattern of rapid, multiple requests and responses that are typical in Ajax applications enable an attacker to look for many vulnerabilities, well-known ports, network locations and so on. The knowledge gained through Ajax fingerprinting can be used to support other attacks, such as XSS.

Other 5 Recently Published CVEs Vulnerabilities

• CVE-2026-2069 – A flaw has been found in ggml-org llama.cpp up to 55abc39. Impacted is the function llama_grammar_advance_stack of the file llama.cpp/src/llama-gra...
• CVE-2026-2068 – A vulnerability was detected in UTT 进取 520W 1.7.7-180627. This issue affects the function strcpy of the file /goform/formSyslogConf. The manipulati...
• CVE-2026-25764 – OpenProject is an open-source, web-based project management software. Prior to versions 16.6.7 and 17.0.3, an HTML injection vulnerability occurs i...
• CVE-2026-25763 – OpenProject is an open-source, web-based project management software. Prior to versions 16.6.7 and 17.0.3, an arbitrary file write vulnerability ex...
• CVE-2026-25760 – Sliver is a command and control framework that uses a custom Wireguard netstack. Prior to 1.6.11, a path traversal in the website content subsystem...