CVE-2026-46678
ADVISORY - githubSummary
Summary
When an application using Pydantic AI opts a URL into force_download='allow-local' (which disables the default block on private/internal IPs), the cloud-metadata blocklist could be bypassed by encoding the metadata IP in an IPv6 transition form (IPv4-mapped IPv6, 6to4, or NAT64). Dual-stack and translated networks route the IPv6 wrapper to the underlying IPv4 endpoint, exposing cloud IAM short-term credentials.
This is an incomplete fix of GHSA-2jrp-274c-jhv3 / CVE-2026-25580. The parent advisory's remediation guaranteed that "cloud metadata endpoints are always blocked, even with allow-local." That guarantee did not hold for IPv6-encoded forms of the metadata IPs.
Severity
Same impact metrics as the parent CVE, but materially narrower attack surface (AC:H instead of AC:L), because exploitation requires the application to have opted into allow-local on a URL influenced by untrusted input.
Who Is Affected
Applications are affected only if they explicitly opt for FileUrl (ImageUrl, AudioUrl, VideoUrl, DocumentUrl) into force_download='allow-local' on a URL that is, or could be, influenced by untrusted input.
Applications are not affected if they use any of the bundled integrations to ingest user input, because they do not propagate force_download from external data:
Agent.to_web/clai webVercelAIAdapterAGUIAdapter/Agent.to_ag_ui
Applications that only download from developer-controlled URLs are not affected.
Remediation
Upgrade to 1.99.0 or later. The cloud-metadata and private-IP blocklists now apply to IPv6 transition forms that route to a blocked IPv4 endpoint (IPv4-mapped IPv6, 6to4, and NAT64 well-known prefix). The blocklists have also been extended to cover additional IANA-reserved IPv4 and IPv6 special-purpose ranges.
Workaround for Unpatched Versions
Avoid passing force_download='allow-local' on any URL that could be influenced by untrusted input. If developers must, resolve the hostname themselves and validate the result against their own metadata blocklist — including IPv6-encoded forms — before constructing the FileUrl.
Credits
Reported by j0hndo.
Common Weakness Enumeration (CWE)
Server-Side Request Forgery (SSRF)
GitHub
2.2
CVSS SCORE
6.8medium| Package | Type | OS Name | OS Version | Affected Ranges | Fix Versions |
|---|---|---|---|---|---|
| pydantic-ai-slim | pypi | - | - | >=1.56.0,<1.99.0 | 1.99.0 |
| pydantic-ai | pypi | - | - | >=1.56.0,<1.99.0 | 1.99.0 |
CVSS:3 Severity and metrics
The CVSS metrics represent different qualitative aspects of a vulnerability that impact the overall score, as defined by the CVSS Specification.
The vulnerable component is bound to the network stack, but the attack is limited at the protocol level to a logically adjacent topology. This can mean an attack must be launched from the same shared physical (e.g., Bluetooth or IEEE 802.11) or logical (e.g., local IP subnet) network, or from within a secure or otherwise limited administrative domain (e.g., MPLS, secure VPN to an administrative network zone). One example of an Adjacent attack would be an ARP (IPv4) or neighbor discovery (IPv6) flood leading to a denial of service on the local LAN segment (e.g., CVE-2013-6014).
A successful attack depends on conditions beyond the attacker's control, requiring investing a measurable amount of effort in research, preparation, or execution against the vulnerable component before a successful attack.
The attacker is unauthorized prior to attack, and therefore does not require any access to settings or files of the vulnerable system to carry out an attack.
The vulnerable system can be exploited without interaction from any user.
An exploited vulnerability can affect resources beyond the security scope managed by the security authority of the vulnerable component. In this case, the vulnerable component and the impacted component are different and managed by different security authorities.
There is a total loss of confidentiality, resulting in all resources within the impacted component being divulged to the attacker. Alternatively, access to only some restricted information is obtained, but the disclosed information presents a direct, serious impact. For example, an attacker steals the administrator's password, or private encryption keys of a web server.
There is no loss of trust or accuracy within the impacted component.
There is no impact to availability within the impacted component.