CVE-2026-107288: Pydantic AI: web_fetch_tool blocked_domains bypass via a hostname the resolver normalizes differently

Vulnerability summary

CVE-2026-107288 is a low (CVSS 3.7) Server-Side Request Forgery vulnerability in pydantic pydantic_ai. EPSS estimates a 0.37% chance of exploitation in the next 30 days, higher than 30.7% of scored vulnerabilities.

Timeline

1. 08-10-2026
   PublishedVulnerability first disclosed.
2. 08-10-2026
   Last modifiedVulnerability information updated.

Description

Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.77.0 until 1.107.6 and 2.44.0, the local web_fetch_tool and the WebFetch local fallback compare blocked_domains entries with a URL hostname before both values are normalized to the form used by getaddrinfo. An attacker-influenced model can use an equivalent IDNA spelling, non-ASCII label separator, case variation, or trailing root label that resolves to a blocked host but does not match the configured string, causing the application to fetch that host with its own privileges. allowed_domains fails closed for unmatched spellings, and private-IP and cloud-metadata protections remain effective. This issue is fixed in versions 1.107.6 and 2.44.0.

• CWE-918Server-Side Request ForgeryThe web server receives a URL or similar request from an upstream component and retrieves the contents of this URL, but it does not sufficiently ensure that the request is being sent to the expected destination.
• CWE-1289Improper Validation of Unsafe Equivalence in InputThe product receives an input value that is used as a resource identifier or other type of reference, but it does not validate or incorrectly validates that the input is equivalent to a potentially-unsafe value.