nday
CVE-2026-65975 GHSA-jpr8-2v3g-wgf9

Pydantic AI AG-UI Adapter: A dangling client-submitted tool call can execute when a trailing message is dropped during `sanitize_messages`

pydantic / pydantic-ai, pydantic-ai-slim

Divulgation : . nday n’a encore aucune trace de correctif ; cela ne veut pas dire qu’il n’en existe pas. Cette vulnérabilité ne figure dans aucune des quatre sources de PoC. Aucune inscription au catalogue KEV de la CISA à ce jour.

why these tags

nday derives these tags by fixed rules from the facts below; NVD, the CNA and other sources do not assign them. Vocabulary and rules: ~/tags.

broken access control
CWE-863, stated by the CNA GitHub_M et NVD, republishing [email protected]
remote, unauthenticated
AV:N, PR:N, UI:N in CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N (CVSS 3.1) from the CNA GitHub_M
network
AV:N in CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N (CVSS 3.1) from the CNA GitHub_M
no authentication required
PR:N in CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N (CVSS 3.1) from the CNA GitHub_M
no user interaction
UI:N in CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N (CVSS 3.1) from the CNA GitHub_M
étape actuelle
DIS
divulgation
CVSS
6.5 MEDIUM v3.1

Cycle de vie

Axe en UTC ; point = minute exacte, barre large = date seule, creux = inconnue ou déduite.
date inconnue
DISdivulguée
07-29T20:06Z
PATcorrectif
aucune donnée
ANAanalyse
aucune donnée
POCPoC
aucune donnée
ITWexploitée
aucune donnée
KEVKEV
aucune donnée

Résumé

Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. In versions 1.88.0 up to but not including 1.107.1 and 2.0.0b1 up to but not including 2.5.0, the UI adapters (AG-UI via Agent.to_ag_ui()/AGUIAdapter, and Vercel AI via VercelAIAdapter) use sanitize_messages to strip unresolved ("dangling") client-submitted tool calls from untrusted message history before it reaches the agent, a defense-in-depth default that prevents the agent from executing tool calls the model never emitted. However, the strip anchored to a message index computed before sanitization ran, so when a trailing client message sanitized to empty and was dropped (for example a client system message under the default manage_system_prompt='server'), a preceding assistant response carrying an unresolved tool call became the new tail and was dispatched without inspection. As a result, a remote client could cause a registered, non-approval server tool to run with client-supplied arguments rather than arguments the model produced. The impact is bounded by what the affected tools do and is most significant for applications that gate tool execution in a model-request hook (before_model_request / after_model_request), since a forged call skips the model turn and bypasses that guardrail; approval-gated tools (requires_approval=True) are not auto-executed by this path. This issue has been fixed in versions 1.107.1 and 2.5.0.

Original text of the CVE record (cvelist), shown as published.

Exploit conditions

What an attack needs, read from the CVSS vector.
Attack vector AV:N
Network: the attacker only needs to reach it over the network
Attack complexity AC:L
Low: no special conditions; the attacker can expect repeatable success
Privileges required PR:N
None: no account or authorization needed
User interaction UI:N
None: no action by anyone other than the attacker

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N

CVSS 3.1 vector from the CVE record, scored by the CNA GitHub_M.

Journal des événements

+DISdivulgation

Fiche CVE publiée par GitHub_M. CVSS 6.5, gravité moyenne.

source cvelist · importé en masse par nday le · brut
{
  "id": 438493,
  "type": "disclosed",
  "source": "cvelist",
  "url": "https://www.cve.org/CVERecord?id=CVE-2026-65975",
  "occurred_at": "2026-07-29T20:06:09.177Z",
  "observed_at": "2026-11-13T05:40:01.626611Z",
  "is_backfill": true,
  "dedupe_key": "",
  "detail": {
    "cvss": {
      "source": "cna",
      "vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N",
      "version": "3.1",
      "base_score": 6.5,
      "base_severity": "MEDIUM"
    },
    "ssvc": {
      "timestamp": "2026-07-30T13:54:02.010071Z",
      "automatable": "yes",
      "exploitation": "none",
      "technical_impact": "partial"
    },
    "state": "PUBLISHED",
    "date_published": "2026-07-29T20:06:09.177Z",
    "assigner_short_name": "GitHub_M"
  }
}

Report an error

report an error or omission

Something on this page wrong or missing? Tell the editors; if they accept it, this page's revision log says so publicly. Upstream data is not changed.

Échanges entre chercheurs

1 chercheur en a parlé ces 7 derniers jours

Publications sur X de chercheurs suivis par nday citant cette CVE, telles que publiées. nday ne les vérifie pas.

  1. 2026-11-20

    Publication de @Snow_Wo1f du 2026-11-20. Connectez-vous pour la lire.

Connectez-vous pour lire 1 publication.

se connecter pour lire

Discussion

0 comments

No comments yet.