GCVE Workshop - 22 September 2026 (14:00-18:00), Luxembourg Before The Vulnopticon Conference - Registration

GHSA-32G3-35G9-WC9G

Vulnerability from github – Published: 2026-06-10 13:38 – Updated: 2026-06-10 13:38
VLAI
Summary
@hulumi/drift: Drift classifier fails open on adapter errors and over-promotes Mixed verdicts
Details

Affected: @hulumi/drift < 1.4.0 — Fixed in: 1.4.0 — Severity: Medium — CWE-755 (Improper Handling of Exceptional Conditions)

Summary

@hulumi/drift runs four adapters that each ask a different question about whether a resource has drifted (Pulumi-state diff, provider-version change, CloudTrail event, etc.). A classifier combines the adapters' answers into a verdict like None / none, ConsoleBreakGlass / high, or Mixed / high, and caches the verdict for 6 hours by default.

Two related bugs from one root cause — the classifier only read each adapter's detected: true/false field and ignored whether the adapter itself succeeded:

  1. Cached "all clear" on adapter failure. When an adapter failed (e.g. transient network error from the Automation API), the classifier read detected: false, concluded "no drift", and cached the verdict as None / none for 6 hours. A single transient failure could mask real console-break-glass mutations for the rest of the window.
  2. Mixed verdicts without real evidence. The Mixed / high and ConsoleBreakGlass / high verdicts (incident severity) could fire on the "the CloudTrail probe round-tripped successfully" signal rather than actual evidence that anything had been changed via the console. Normal provider-API churn could end up falsely escalated to incident severity.

Impact

Consumers running drift detection in CI / cron could see transient adapter failures silently cached as "all clear" — masking real attacks for up to six hours — or see ordinary provider-version churn falsely promoted to incident severity. Either way, the verdict source was unreliable for downstream incident workflows that gate on it.

Patches

Upgrade to @hulumi/drift@1.4.0. Classifier-only fix (the TLA+-verified 6-row verdict matrix is byte-identical):

  • adapter failures now fail closed to Unknown / low, and degraded verdicts are not written to the cache;
  • the Mixed / ConsoleBreakGlass promotion now requires real CloudTrail event evidence rather than probe liveness.

Workarounds

Setting options.minConfidence: "medium" on the classifier call prevents the degraded None / none from being cached (it doesn't meet the threshold), partially mitigating case (1). No workaround for case (2).

Resources

  • PR #178 (Cluster D); regression tests in packages/drift/tests/classifier-fail-closed.test.ts.
Severity
8.4 (High)
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:N/VI:H/VA:L/SC:N/SI:H/SA:L
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "npm",
        "name": "@hulumi/drift"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "1.4.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-48036"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-755"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-06-10T13:38:50Z",
    "nvd_published_at": null,
    "severity": "HIGH"
  },
  "details": "**Affected:** `@hulumi/drift` `\u003c 1.4.0` \u2014 **Fixed in:** `1.4.0` \u2014 **Severity:** Medium \u2014 **CWE-755 (Improper Handling of Exceptional Conditions)**\n\n#### Summary\n\n`@hulumi/drift` runs four adapters that each ask a different question about whether a resource has drifted (Pulumi-state diff, provider-version change, CloudTrail event, etc.). A classifier combines the adapters\u0027 answers into a verdict like `None / none`, `ConsoleBreakGlass / high`, or `Mixed / high`, and caches the verdict for 6 hours by default.\n\nTwo related bugs from one root cause \u2014 the classifier only read each adapter\u0027s `detected: true/false` field and ignored whether the adapter itself succeeded:\n\n1. **Cached \"all clear\" on adapter failure.** When an adapter failed (e.g. transient network error from the Automation API), the classifier read `detected: false`, concluded \"no drift\", and cached the verdict as `None / none` for 6 hours. A single transient failure could mask real console-break-glass mutations for the rest of the window.\n2. **Mixed verdicts without real evidence.** The `Mixed / high` and `ConsoleBreakGlass / high` verdicts (incident severity) could fire on the \"the CloudTrail probe round-tripped successfully\" signal rather than actual evidence that anything had been changed via the console. Normal provider-API churn could end up falsely escalated to incident severity.\n\n#### Impact\n\nConsumers running drift detection in CI / cron could see transient adapter failures silently cached as \"all clear\" \u2014 masking real attacks for up to six hours \u2014 or see ordinary provider-version churn falsely promoted to incident severity. Either way, the verdict source was unreliable for downstream incident workflows that gate on it.\n\n#### Patches\n\nUpgrade to `@hulumi/drift@1.4.0`. Classifier-only fix (the TLA+-verified 6-row verdict matrix is byte-identical):\n\n- adapter failures now fail closed to `Unknown / low`, and degraded verdicts are not written to the cache;\n- the `Mixed` / `ConsoleBreakGlass` promotion now requires real CloudTrail event evidence rather than probe liveness.\n\n#### Workarounds\n\nSetting `options.minConfidence: \"medium\"` on the classifier call prevents the degraded `None / none` from being cached (it doesn\u0027t meet the threshold), partially mitigating case (1). No workaround for case (2).\n\n#### Resources\n\n- [PR #178](https://github.com/kerberosmansour/hulumi/pull/178) (Cluster D); regression tests in `packages/drift/tests/classifier-fail-closed.test.ts`.",
  "id": "GHSA-32g3-35g9-wc9g",
  "modified": "2026-06-10T13:38:50Z",
  "published": "2026-06-10T13:38:50Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/kerberosmansour/hulumi/security/advisories/GHSA-32g3-35g9-wc9g"
    },
    {
      "type": "WEB",
      "url": "https://github.com/kerberosmansour/hulumi/pull/178"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/kerberosmansour/hulumi"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:N/VI:H/VA:L/SC:N/SI:H/SA:L",
      "type": "CVSS_V4"
    }
  ],
  "summary": "@hulumi/drift: Drift classifier fails open on adapter errors and over-promotes Mixed verdicts"
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





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.


Detection rules are retrieved from Rulezet.


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.