GHSA-5JJJ-9XC3-RM56
Vulnerability from github – Published: 2026-10-07 20:24 – Updated: 2026-10-07 20:24Summary
hydra-optuna-sweeper accepted a configuration-controlled dotted path in
hydra.sweeper.custom_search_space, resolved it with the lower-level
hydra.utils.get_method() API, and later invoked the returned callable.
Hydra's object-lookup helpers intentionally trust their input and do not apply
the execution policy used by instantiate() and logging configuration. As a
result, untrusted Optuna multirun configuration could select installed Python
code for execution in the Hydra controller process.
Impact
Exploitation requires control of an application's Optuna sweep configuration or command-line overrides and an importable callable in the application's environment. Selected code runs with the privileges of the application.
On affected Hydra 1.4 development releases, this path bypasses an execution whitelist provided by trusted application code. Hydra 1.3 has no execution-whitelist security boundary; there, the same change restores the legacy blocklist as defense in depth.
Fix
The Optuna sweeper now resolves the configured callback through
hydra.utils.instantiate() as a partial callable. On Hydra 1.4, the active
execution whitelist therefore authorizes the callback before use. On Hydra
1.3, the legacy blocklist is applied as defense in depth.
The public get_object(), get_method(), get_static_method(), and
get_class() helpers remain lower-level trusted-input APIs. Their documentation
now warns against passing values derived from untrusted configuration and
directs config-driven construction to instantiate().
Affected versions
- Package:
hydra-optuna-sweeper - Stable affected range:
>= 1.2.0, < 1.3.0 - Stable fixed version:
1.3.0 - Development affected range:
>= 1.4.0.dev4, < 1.4.0.dev10 - Development fixed version:
1.4.0.dev10
The feature was introduced in 1.2.0 and is absent from the 1.1 release line.
The 1.4 development line was affected from 1.4.0.dev4 through 1.4.0.dev9.
Workaround
Treat Optuna sweep configuration and command-line overrides as trusted. Do not
set custom_search_space from externally controlled input or allow untrusted
users to place modules on the application import path.
Reported by zx (GitHub: @manus-pi).
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "hydra-optuna-sweeper"
},
"ranges": [
{
"events": [
{
"introduced": "1.2.0"
},
{
"fixed": "1.3.0"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "PyPI",
"name": "hydra-optuna-sweeper"
},
"ranges": [
{
"events": [
{
"introduced": "1.4.0.dev4"
},
{
"fixed": "1.4.0.dev10"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-106440"
],
"database_specific": {
"cwe_ids": [
"CWE-470"
],
"github_reviewed": true,
"github_reviewed_at": "2026-10-07T20:24:02Z",
"nvd_published_at": "2026-10-06T19:18:12Z",
"severity": "HIGH"
},
"details": "## Summary\n\n`hydra-optuna-sweeper` accepted a configuration-controlled dotted path in\n`hydra.sweeper.custom_search_space`, resolved it with the lower-level\n`hydra.utils.get_method()` API, and later invoked the returned callable.\n\nHydra\u0027s object-lookup helpers intentionally trust their input and do not apply\nthe execution policy used by `instantiate()` and logging configuration. As a\nresult, untrusted Optuna multirun configuration could select installed Python\ncode for execution in the Hydra controller process.\n\n## Impact\n\nExploitation requires control of an application\u0027s Optuna sweep configuration\nor command-line overrides and an importable callable in the application\u0027s\nenvironment. Selected code runs with the privileges of the application.\n\nOn affected Hydra 1.4 development releases, this path bypasses an execution\nwhitelist provided by trusted application code. Hydra 1.3 has no execution-whitelist\nsecurity boundary; there, the same change restores the legacy blocklist as\ndefense in depth.\n\n## Fix\n\nThe Optuna sweeper now resolves the configured callback through\n`hydra.utils.instantiate()` as a partial callable. On Hydra 1.4, the active\nexecution whitelist therefore authorizes the callback before use. On Hydra\n1.3, the legacy blocklist is applied as defense in depth.\n\nThe public `get_object()`, `get_method()`, `get_static_method()`, and\n`get_class()` helpers remain lower-level trusted-input APIs. Their documentation\nnow warns against passing values derived from untrusted configuration and\ndirects config-driven construction to `instantiate()`.\n\n## Affected versions\n\n- Package: `hydra-optuna-sweeper`\n- Stable affected range: `\u003e= 1.2.0, \u003c 1.3.0`\n- Stable fixed version: `1.3.0`\n- Development affected range: `\u003e= 1.4.0.dev4, \u003c 1.4.0.dev10`\n- Development fixed version: `1.4.0.dev10`\n\nThe feature was introduced in `1.2.0` and is absent from the 1.1 release line.\nThe 1.4 development line was affected from 1.4.0.dev4 through 1.4.0.dev9.\n\n## Workaround\n\nTreat Optuna sweep configuration and command-line overrides as trusted. Do not\nset `custom_search_space` from externally controlled input or allow untrusted\nusers to place modules on the application import path.\n\nReported by zx (GitHub: @manus-pi).",
"id": "GHSA-5jjj-9xc3-rm56",
"modified": "2026-10-07T20:24:02Z",
"published": "2026-10-07T20:24:02Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/hydra-ecosystem/hydra/security/advisories/GHSA-5jjj-9xc3-rm56"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-106440"
},
{
"type": "WEB",
"url": "https://github.com/hydra-ecosystem/hydra/commit/0dd18084589a3d3e577d1f1a8a48fb485c94a5e6"
},
{
"type": "WEB",
"url": "https://github.com/hydra-ecosystem/hydra/commit/4720dfca2bde27fa140ec287a05709668fbdf168"
},
{
"type": "PACKAGE",
"url": "https://github.com/hydra-ecosystem/hydra"
},
{
"type": "WEB",
"url": "https://github.com/hydra-ecosystem/hydra/releases/tag/v1.3.7"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "Hydra: Optuna custom_search_space can resolve and execute untrusted callables via get_method"
}
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.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
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.