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.

An attacker targeting an AI engineering team discovers via LinkedIn or GitHub that developers are using Pydantic AI to build an internal RAG assistant with an Agent.to_web() UI. The attacker crafts a URL to the internal app with a 'version' parameter containing path traversal sequences pointing to attacker-controlled HTML hosted on the legitimate CDN used by Pydantic AI. The URL is sent as a Slack DM ('hey, can you check if this agent response looks right to you?') or embedded in a GitHub issue comment on the target's repo. When the developer opens the link, the attacker's JavaScript executes in the context of the Pydantic AI chat interface, silently POSTing the full conversation history — including any API keys, internal system prompts, or proprietary data typed into the chat — to an attacker-controlled endpoint. The developer sees nothing unusual.