Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessIdentify app with allow-local opt-in on IPv6 host
-
DeliverySupply zone-scoped IPv6 metadata URL (fd00:ec2::254%251)
-
ExploitBypass blocklist via zone-id equality mismatch
-
ExecutionReach cloud metadata endpoint
-
ImpactExfiltrate IAM short-term credentials
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A01:2025 - Broken Access Control
ASVS 5.0
V4 - API and Web Service URL allowlisting, SSRF prevention
MITRE ATT&CK
2 techniques across 2 tactics: Initial Access, Credential Access The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Exploitation requires ALL of the following concrete preconditions, each confirmed by the vendor advisory: (1) the application must explicitly opt a URL into local network access via either a FileUrl (ImageUrl/AudioUrl/VideoUrl/DocumentUrl) with force_download='allow-local' or web_fetch_tool(allow_local_urls=True) - both OFF BY DEFAULT; (2) that URL must be influenceable by untrusted input (attacker-controlled, or model-chosen in the web_fetch_tool path); (3) the runtime must be an IPv6-enabled cloud environment with a reachable metadata service (e.g. AWS EC2/EKS with IPv6, GCP IPv6-only instances, Scaleway), since only the IPv6 metadata endpoints were reachable via zone-identifier smuggling. Limiting factors: applications using bundled ingestion integrations (e.g. Agent.to_web / clai web) are NOT affected via the FileUrl path because they do not propagate force_download from external data; IPv4, private-range, and transition-form metadata checks were never bypassable. The specific bypass is appending an IPv6 zone identifier (e.g. fd00:ec2::254%251) to a blocked metadata address. |
| Impact | This is a genuinely lower-priority issue despite its CWE-918 SSRF classification, and the signals are internally consistent rather than conflicting. The CVSS 3.1 vector (AV:N/AC:H/PR:N/UI:N/S:C/C:H/I:N/A:N, 6.8 Medium) correctly captures the tension: although the attacker needs no privileges on the target (PR:N) and no user interaction (UI:N), the high attack complexity (AC:H) reflects a stack of required preconditions that the vendor advisory confirms - the application must explicitly opt an attacker-influenced URL into local network access (force_download='allow-local' or web_fetch_tool with allow_local_urls=True), both of which are DISABLED BY DEFAULT, and the environment must be IPv6-enabled (e.g. AWS EC2/EKS with IPv6, GCP IPv6-only, Scaleway). The scope change (S:C) and C:H are justified because success exposes cloud IAM short-term credentials from the metadata service, crossing a trust boundary. There is no EPSS score, KEV listing, or POC in the provided data - no public exploit identified at time of analysis - and this absence is consistent with a defense-in-depth hardening fix rather than an actively exploited flaw. Notably, this is explicitly an INCOMPLETE FIX of prior advisories (GHSA-cqp8-fcvh-x7r3/CVE-2026-46678 and GHSA-cg7w-rg45-pc59/CVE-2026-48782, themselves follow-ups to CVE-2026-25580), which raises the realistic likelihood that security-conscious operators who already allow local URLs are the exposed population. Net assessment: real but narrow; prioritize patching only where the opt-in local-access settings are actually enabled in an IPv6 environment. |
| Exploit ScenarioDerived | An application has enabled local-URL access (force_download='allow-local' or web_fetch_tool(allow_local_urls=True)) and runs on an IPv6-enabled cloud instance. An attacker who can influence a fetched URL - or the LLM-chosen URL in the web_fetch_tool case - supplies a zone-scoped IPv6 metadata address such as http://[fd00:ec2::254%251]/latest/meta-data/iam/security-credentials/; the blocklist comparison fails because Python folds the zone id into IPv6Address equality, while the kernel ignores the zone for the non-link-local destination and reaches the metadata service, returning cloud IAM credentials. No public exploit was identified at time of analysis. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 08, 2026
Server-side request forgery in Pydantic AI's URL-fetching guards lets an application that has explicitly opted a URL into local-network access reach IPv6 cloud metadata endpoints such as fd00:ec2::254 and read short-lived cloud IAM credentials, because appending an RFC 4007 zone identifier (for example fd00:ec2::254%251) makes the address unequal to its blocklist entry while the kernel silently ignores the zone for a non-link-local destination and routes the request anyway. Affected releases are pydantic-ai and pydantic-ai-slim at 1.56.0 up to but not including 1.107.6, and 2.0.0b1 up to but not including 2.44.0; exploitation additionally requires that the URL be attacker-influenced or model-chosen, that one of the two opt-ins be enabled (FileUrl with force_download='allow-local', or web_fetch_tool(allow_local_urls=True) - both off by default), and an IPv6-enabled environment with a reachable metadata service. Vendor-released patches are available in versions 1.107.6 and 2.44.0 (upstream fix tracked via pull requests 8401 and 8402); no public exploit code was identified at time of analysis and there is no confirmed active exploitation (CISA KEV).
RemediationAI
Vendor-released patch: upgrade to pydantic-ai / pydantic-ai-slim 1.107.6 (1.x line) or 2.44.0 (2.x line), which strip IPv6 zone identifiers before any blocklist comparison (upstream pull requests 8401 and 8402, commits 02157e1b87bd45d3f2e111ce07afdf89f9fb0e5b and 4da70591460f51a8c4f128eaeef70a33340dbd55). If immediate upgrade is not possible, the most effective compensating control is to not opt into local access at all: leave FileUrl/force_download at its default and do not call web_fetch_tool(allow_local_urls=True) - this fully removes exposure on both vulnerable paths, with the trade-off that the application loses the ability to fetch legitimately local or intranet resources through those APIs. Where local access is genuinely required, never pass untrusted or model-chosen URLs into those paths: hard-code the URL, or validate attacker-influenced URLs against an explicit hostname/IP allowlist that rejects any address containing a '%' zone identifier before it reaches Pydantic AI; the trade-off is added application logic and the risk of a permissive allowlist reintroducing SSRF. Environment-level mitigations reduce the blast radius rather than the bug: enforce IMDSv2 with a short hop limit or disable the IPv6 metadata endpoint on AWS (and equivalent link-local restrictions on GCP/Scaleway) so a successful request cannot retrieve IAM credentials, accepting that this does not prevent the outbound SSRF itself; and apply outbound egress filtering that blocks the IPv6 metadata prefixes fd00:ec2::/64, fd20:ce::/64 and fd00:42::42 from the workload, which may break other legitimate link-local or private IPv6 traffic. Because the two opt-in settings are disabled by default and only the IPv6 metadata endpoints were reachable, confirm exposure by checking whether your deployment combines an enabled opt-in, untrusted URL input, and an IPv6-enabled cloud network before treating this as urgent.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 4 packages in pypi, euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
The flaw sits in Pydantic AI's SSRF guard module (pydantic_ai/_ssrf.py), which maintains a frozenset of blocked cloud metadata endpoints and, for opt-in local access, is supposed to unconditionally refuse requests to them. Root cause is CWE-918 (Server-Side Request Forgery) compounded by a Python semantics edge case: ipaddress.IPv6Address folds an RFC 4007 zone/scope identifier into equality and hashing, so a zone-scoped spelling of a blocked address is not a member of the frozenset, while the operating system treats the zone as meaningful only for a link-local destination and ignores it otherwise. The vulnerability is therefore an incomplete fix: it is a follow-up to GHSA-cqp8-fcvh-x7r3 / CVE-2026-46678 and GHSA-cg7w-rg45-pc59 / CVE-2026-48782, which were themselves follow-ups to CVE-2026-25580, whose remediation advertised a guarantee that cloud metadata endpoints are always blocked even with local access allowed. Notably, the private-range checks and the IPv4 and IPv4-transition-form (IPv4-mapped, IPv4-compatible, 6to4, NAT64, ISATAP) metadata checks were never bypassable, because they compare by network containment and by packed bytes respectively rather than by set membership - only the IPv6 metadata list was exposed. The upstream patch introduces a _parse_ip helper that strips the scope_id before any comparison, so address classification is unaffected by the zone while the connection still carries it (link-local destinations with zones remain reachable under allow-local). Only IPv6 metadata endpoints were reachable through this path, so an IPv6-enabled runtime - AWS EC2/EKS with IPv6, GCP IPv6-only instances, or Scaleway - is required.
Know when Pydantic Ai gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:N/AC:H/PR:N/UI:N/S:C/C:H/I:N/A:N
Attacker needs no target privileges (PR:N) but the opt-in local-access config plus IPv6 environment make AC:H; leaking IAM creds from the metadata service crosses scope (S:C) with C:H and no integrity/availability impact.
Primary rating from Vendor (GitHub_M).
CVSS VectorVendor: GitHub_M
Lifecycle Timeline
1DescriptionCVE.org
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.56.0 until 1.107.6 and 2.44.0, applications that opt attacker-influenced URLs into local network access through FileUrl with force_download='allow-local' or web_fetch_tool with allow_local_urls=True can bypass the cloud-metadata blocklist by appending an IPv6 zone identifier to an IPv6 metadata address. IPv6Address equality and hashing include the zone identifier, so the blocklist comparison fails even though the network stack ignores the zone on a non-link-local destination and reaches the metadata service, potentially exposing cloud IAM credentials. The opt-in settings are disabled by default, and the issue requires an IPv6-enabled environment. This issue is fixed in versions 1.107.6 and 2.44.0.
Blast Radius
ecosystem impact- 2 pypi packages depend on pydantic-ai (2 direct, 0 indirect)
- 12 pypi packages depend on pydantic-ai-slim (10 direct, 2 indirect)
Ecosystem-wide dependent count for version 1.56.0 and other introduced versions.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Server-Side Request Forgery exploits applications that fetch remote resources based on user-supplied URLs.
More in Pydantic Ai
View allPydantic AI versions 0.0.26 through 1.55.x contain a server-side request forgery vulnerability in URL download functiona
Cross-site request forgery in Pydantic AI's development web chat UI allows any website a developer happens to visit whil
Concurrency slot exhaustion in Pydantic AI 2.10.0 through 2.52.x lets streamed model requests permanently occupy slots i
Server-Side Request Forgery in Pydantic AI (versions 1.56.0-1.101.0, 2.0.0b1, 2.0.0b2) allows unauthenticated network at
Server-Side Request Forgery in Pydantic AI's UI adapter layer (versions 1.65.0-1.105.0 and 2.0.0b1-2.0.0b5) allows unaut
Pydantic AI, a Python framework for building Generative AI agents, buffers complete HTTP response bodies before applying
Denial of service in Pydantic AI (pydantic-ai and pydantic-ai-slim) versions 1.77.0 through 1.107.5 and 2.0.0b1 through
Denial of service in Pydantic AI's local web_fetch_tool and WebFetch local fallback allows an attacker-controlled HTML p
Tool call injection in Pydantic AI's UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdap
DNS rebinding against the Pydantic AI development web chat server (Agent.to_web() and `clai web`) lets a hostile website
Pydantic AI versions 1.34.0 through 1.50.x contain a path traversal vulnerability in the web UI that allows unauthentica
A domain-denylist bypass in Pydantic AI's local web-fetch tooling (web_fetch_tool and the WebFetch local fallback) lets
EUVD-2026-95077
GHSA-vmxc-h2x2-jmf3
NVD
CVE.org
EPSS
Vulnrichment
Exploit-DB