ShieldBreak / RoguePlanet · case reference

What changes between the check and the use?

A security engine checks a file or path, then uses it with elevated privileges. This case compares three ways the value used can differ from the value checked.

Two variations come from the ShieldBreak model on the homepage: changing the file's contents, and redirecting an object path. The third comes from the related RoguePlanet case and uses a folder junction.

Prepared walkthrough Archive: 5 September 2026

Reference material for the homepage ShieldBreak walkthrough.

The inferred check-to-use group
Check a valueSubstitute bytes or a pathUse it with privilege

Open a variation to read its conditions and sources.

The three variations
VariationWhat differs between the check and the use
Content substitution · M3The engine checks one version of the file and later reads different contents during cleanup.
Object-path redirection · M4An Object Manager link (a link in the Windows kernel object namespace that redirects one object path to another) sends the engine's privileged write to a different destination.
Junction-path redirection · N3A folder junction (a Windows directory that redirects to another directory) changes the destination between the engine's path check and its write.
Archived 5 September 2026 · requirements inferred17 source records · 13 excerpts · 1 full-text label · 1 repository description · 2 missingMethod and limits

01 / Understand

The value the engine checked and the value it later used

The engine checks the file or path at one moment and uses it with elevated privileges at a later one. The selected variation shows the value at each moment and what changed between them.

Shared requirements

  • The attacker can run code as a normal user.
  • A privileged security engine is working on an attacker-controlled path.

Change the value after a check and before privileged use.

Requirement basis

These requirements were authored from sources and remain inferred; no perturbation tests were run. The shared grouping is a derived model conclusion.

Content substitution

Swap the file contents while the security engine is cleaning it up.

Record scope

The served bytes and their revalidation gap are the record’s required variation beyond the check-to-use group.

Object-path redirection

Redirect where the security engine's privileged write lands.

Record scope

Object Manager link rebinding and its revalidation gap are the record’s required variation beyond the check-to-use group.

Junction-path redirection

Swap the folder link at the exact moment the security engine is using it.

Record scope

NTFS junction rebinding and its revalidation gap are the record’s required variation beyond the check-to-use group.

Follow the implication for a candidate fix →

02 / Act

Where a fix would have to act

A fix has to stop the checked object from changing before the engine uses it, or make the privileged operation act on the same object that passed the check. A second check alone can leave another window for a change before use. In the selected variation, the point between the check and the use is where either fix would apply.

Proposed fix · Not tested

Define the boundary a candidate fix actually closes

Intervention principle
Keep the checked object stable until use, or bind the privileged operation to the object that passed the check. A second check alone can leave another race window.
Defensive capability
Revalidation and race-resistant handling of content and resolved paths in privileged security-engine operations.

Proposed scope review · implementation status: intended

Test content substitution

The security engine does not re-check the contents before acting on them.

Formal condition under review

contentRevalidation = absent

Hold the other modeled conditions fixed, vary this condition, and observe whether the privileged effect still follows. This is a test proposal; no result is supplied.

Test object-path redirection

The security engine does not re-check the path before it writes.

Formal condition under review

objectPathRevalidation = absent

Hold the other modeled conditions fixed, vary this condition, and observe whether the privileged effect still follows. This is a test proposal; no result is supplied.

Test junction-path redirection

The security engine does not re-check the path between checking it and writing to it.

Formal condition under review

junctionPathRevalidation = absent

Hold the other modeled conditions fixed, vary this condition, and observe whether the privileged effect still follows. This is a test proposal; no result is supplied.

The model supplies a place to test the proposed capability. A result on one variation remains scoped to that variation until evidence supports a broader conclusion.

Review the assumptions before claiming a result →

03 / Explore

Three variations of one pattern

In all three records a value changes between the engine's check and its use (the time-of-check to time-of-use pattern). What changes differs: the file's contents in M3 (content substitution), an object path in M4 (object-path redirection), or a junction path in N3 (junction-path redirection). Matching conditions identify a possible comparison; applying one fix to both also requires evidence about where and how it acts. Compare the three records to find those matches.

Swap the file contents during cleanup

Inferred requirements

Swap the file contents while the security engine is cleaning it up.

Needs

  • The attacker can run code as a normal user.
  • The attacker's process decides what the security engine reads from that file.
  • The security engine is scanning the attacker's path as SYSTEM.
  • The security engine does not re-check the contents before acting on them.

Changes

  • Hand the security engine a file it will flag, so it starts cleaning it up.
  • Hand it different contents when it reads the file again during cleanup.

Leaves behind

  • The security engine is now working on bytes the attacker chose, with SYSTEM rights.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • hydrationCallbackOwner = adversary
  • scannerAttention = onAdversaryPath
  • contentRevalidation = absent

M · Mechanism

  • serve detected test content on the first hydration, so the engine enters remediation
  • serve different content on the restart hydration that the remediation read triggers

Q · Postconditions

  • privilegedProcessInput = adversaryControlled

Source trail for this record: "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only. · Sheel, Vamika. "CVE-2026-69414 ShieldBreak Zero-Day: No Patch, and CISA BOD 26-04 Gives You 14 Days." Qualys Blog, 25 August 2026, updated 3 September 2026.

Inspect this formal record →

Redirect the privileged write through an object path

Inferred requirements

Redirect where the security engine's privileged write lands.

Needs

  • The attacker can run code as a normal user.
  • The security engine is working on the attacker's path as SYSTEM.
  • A normal user can create Object Manager symbolic links in their own namespace.
  • The security engine does not re-check the path before it writes.

Changes

  • Change where the link points, after the security engine has checked the path and before it writes.

Leaves behind

  • The security engine's SYSTEM-level write lands wherever the attacker chose.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • objectSymlinkCreation ⊒ permittedToUser
  • objectPathRevalidation = absent

M · Mechanism

  • rebind Object Manager symbolic links between the engine's path resolution and its use

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen

Source trail for this record: "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only. · "ShieldBreak: A Local Privilege Escalation Path Through Microsoft Defender." CyberMaxx, August 2026. · "CVE-2026-69414 Defender Flaw Has No Fix or Affected Builds." Windows Forum, 3 September 2026.

Inspect this formal record →

Redirect the privileged write through a folder junction

Inferred requirements

Swap the folder link at the exact moment the security engine is using it.

Needs

  • The attacker can run code as a normal user.
  • The security engine is working on the attacker's path as SYSTEM.
  • A normal user can create NTFS junctions.
  • The security engine does not re-check the path between checking it and writing to it.

Changes

  • Repoint the junction in the gap between the security engine's check and its write. Retry until the timing lands.

Leaves behind

  • The security engine's SYSTEM-level write lands wherever the attacker chose.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • junctionCreation ⊒ permittedToUser
  • junctionPathRevalidation = absent

M · Mechanism

  • rebind an NTFS junction between the engine's path check and its quarantine write, repeating until the race is won

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen

Source trail for this record: "CVE-2026-50656: Microsoft Defender RoguePlanet Zero-Day Enables Local Privilege Escalation on Fully Patched Windows 10 and 11 Systems." Rescana, 18 June 2026. · "Microsoft fixed Defender flaw RoguePlanet (CVE-2026-50656)." Security Affairs, 9 July 2026. Source for the Windows Server mounting limitation. · "CVE-2026-50656 'RoguePlanet': Microsoft Defender Malware Protection Engine Privilege Escalation to SYSTEM." Threat-Modeling.com, 9 July 2026. Source for engine version 1.1.26060.3008.

Inspect this formal record →

04 / Challenge

What would overturn the model

Choose a requirement or a link between two actions. Which observation would show it unnecessary, incomplete, or wrong? That observation is the next test. A fix tested against one variation leaves its effect on the other two open.

Selected-record requirements · inferred

Would content substitution still work without one of these conditions?

Conditions in the selected record

Needs

  • The attacker can run code as a normal user.
  • The attacker's process decides what the security engine reads from that file.
  • The security engine is scanning the attacker's path as SYSTEM.
  • The security engine does not re-check the contents before acting on them.

Changes

  • Hand the security engine a file it will flag, so it starts cleaning it up.
  • Hand it different contents when it reads the file again during cleanup.

Leaves behind

  • The security engine is now working on bytes the attacker chose, with SYSTEM rights.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • hydrationCallbackOwner = adversary
  • scannerAttention = onAdversaryPath
  • contentRevalidation = absent

M · Mechanism

  • serve detected test content on the first hydration, so the engine enters remediation
  • serve different content on the restart hydration that the remediation read triggers

Q · Postconditions

  • privilegedProcessInput = adversaryControlled

Each listed precondition is required by the authored record. None has been established by a perturbation test. Vary one while holding the others fixed; a contradictory result should revise this record and the comparisons that depend on it.

Selected-record requirements · inferred

Would object-path redirection still work without one of these conditions?

Conditions in the selected record

Needs

  • The attacker can run code as a normal user.
  • The security engine is working on the attacker's path as SYSTEM.
  • A normal user can create Object Manager symbolic links in their own namespace.
  • The security engine does not re-check the path before it writes.

Changes

  • Change where the link points, after the security engine has checked the path and before it writes.

Leaves behind

  • The security engine's SYSTEM-level write lands wherever the attacker chose.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • objectSymlinkCreation ⊒ permittedToUser
  • objectPathRevalidation = absent

M · Mechanism

  • rebind Object Manager symbolic links between the engine's path resolution and its use

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen

Each listed precondition is required by the authored record. None has been established by a perturbation test. Vary one while holding the others fixed; a contradictory result should revise this record and the comparisons that depend on it.

Selected-record requirements · inferred

Would junction-path redirection still work without one of these conditions?

Conditions in the selected record

Needs

  • The attacker can run code as a normal user.
  • The security engine is working on the attacker's path as SYSTEM.
  • A normal user can create NTFS junctions.
  • The security engine does not re-check the path between checking it and writing to it.

Changes

  • Repoint the junction in the gap between the security engine's check and its write. Retry until the timing lands.

Leaves behind

  • The security engine's SYSTEM-level write lands wherever the attacker chose.
Formal preconditions, mechanism and postconditions

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • junctionCreation ⊒ permittedToUser
  • junctionPathRevalidation = absent

M · Mechanism

  • rebind an NTFS junction between the engine's path check and its quarantine write, repeating until the race is won

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen

Each listed precondition is required by the authored record. None has been established by a perturbation test. Vary one while holding the others fixed; a contradictory result should revise this record and the comparisons that depend on it.

Shared across the three variations

Do the inferred requirements survive controlled variation?

The archived model was reconstructed from reports. No test removed or varied a required element while holding the rest of the setup fixed.

If challenged: Keep the requirements and the check-to-use group provisional. A contradictory experiment should revise the required set and recompute dependent cores and edges.

Shared across the three variations

How much of the registry was actually available?

Only one reference carries an archived full-text retrieval label; thirteen carry search-excerpt labels, one has repository-description-only access, and two were not retrieved. The excerpt for reference 17 also records refused direct retrieval.

If challenged: Source count and publisher diversity cannot establish independent confirmation, full evidence access or current validity.

Shared across the three variations

What would establish a real defensive result?

A model edge carries at least one precondition clause. It does not prove all later clauses hold, nor that a deployment can reach a preventing state.

If challenged: A reachability or protection claim needs a scoped posture and tests beyond this type-level graph.

Related research · outside the curated selection

Does the execution step (M5) need both suppliers?

The archived execution step (M5) record requires both supplier clauses. A residual experiment would vary privilegedProcessInput and observe whether SYSTEM execution still follows with the remaining conditions held.

If challenged: If that clause proved incidental, the archive says the execution step (M5) and the related execution step (N4) would have equal required sets and the execution group (C3) would need withdrawal as a distinct multi-record estimate. That outcome is not claimed.

Shared across the three variations

Can the formal bindings be reconciled?

The specification corpus used by the archive differs from the ontology’s recorded specification pin.

If challenged: Keep every archived ontology binding provisional until the pins are reconciled against authoritative source material.

Prepare a challenge or alternative analysis →

05 / Verify

What the sources establish

Each claim lists its sources and what each one supports. The badge separates a reported claim from an inferred one, and the source entry shows what material the archive retained.

Archive record
Retained in the supplied archive; inspect its evidence basis separately.
Inferred
A reconstructed model conclusion.
Unassessed
Not assessed by this snapshot.

Content substitution · record evidence

Swap the file contents while the security engine is cleaning it up.

Requirement basis: inferred; no perturbation · provenance: asserted

Curated archival support: Content substitution · formal source

Source trail for this record: "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only. · Sheel, Vamika. "CVE-2026-69414 ShieldBreak Zero-Day: No Patch, and CISA BOD 26-04 Gives You 14 Days." Qualys Blog, 25 August 2026, updated 3 September 2026.

Object-path redirection · record evidence

Redirect where the security engine's privileged write lands.

Requirement basis: inferred; no perturbation · provenance: asserted

Curated archival support: Object-path redirection · formal source

Source trail for this record: "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only. · "ShieldBreak: A Local Privilege Escalation Path Through Microsoft Defender." CyberMaxx, August 2026. · "CVE-2026-69414 Defender Flaw Has No Fix or Affected Builds." Windows Forum, 3 September 2026.

Junction-path redirection · record evidence

Swap the folder link at the exact moment the security engine is using it.

Requirement basis: inferred; no perturbation · provenance: asserted

Curated archival support: Junction-path redirection · formal source

Source trail for this record: "CVE-2026-50656: Microsoft Defender RoguePlanet Zero-Day Enables Local Privilege Escalation on Fully Patched Windows 10 and 11 Systems." Rescana, 18 June 2026. · "Microsoft fixed Defender flaw RoguePlanet (CVE-2026-50656)." Security Affairs, 9 July 2026. Source for the Windows Server mounting limitation. · "CVE-2026-50656 'RoguePlanet': Microsoft Defender Malware Protection Engine Privilege Escalation to SYSTEM." Threat-Modeling.com, 9 July 2026. Source for engine version 1.1.26060.3008.

Shared across the three variations

Three records share a check-to-use pattern

Inferred

The records imply it. The claim details retain its assumptions and supporting records.

Open claim details
4 cited sources

In the archived model, content substitution, object-path redirection and junction-path redirection require user-level execution and a privileged security engine working on an attacker-controlled path. Each changes the value used after a check. This is the estimated check-to-use group pattern; the bytes, object path and junction path remain different variations.

Claim details and sources
  • The check-to-use group has provisional identity. Its three modeled members are an observed population, not the definition of membership.
  • No requirement or invariant was established by perturbation.
  • Both incidents trace to one reported researcher; secondary agreement is not independent confirmation.

Curated archival support: Shared check-to-use group · Content substitution · formal source · Object-path redirection · formal source · Junction-path redirection · formal source

External references for this claim: "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only. · "CVE-2026-50656: Microsoft Defender RoguePlanet Zero-Day Enables Local Privilege Escalation on Fully Patched Windows 10 and 11 Systems." Rescana, 18 June 2026. · "Microsoft fixed Defender flaw RoguePlanet (CVE-2026-50656)." Security Affairs, 9 July 2026. Source for the Windows Server mounting limitation. · "CVE-2026-50656 'RoguePlanet': Microsoft Defender Malware Protection Engine Privilege Escalation to SYSTEM." Threat-Modeling.com, 9 July 2026. Source for engine version 1.1.26060.3008.

Provenance: derived

Shared across the three variations

A result confined to one variation has a confined scope

Inferred

The records imply it. The claim details retain its assumptions and supporting records.

Open claim details
3 cited sources

A change aimed only at junction-path revalidation does not, from the model alone, establish protection for the byte substitution or Object Manager path variation. A wider transfer claim needs evidence that the change acts on the check-to-use group’s invariant and satisfies the relevant scope conditions.

Claim details and sources
  • This is a structural comparison, not a statement about current patches, affected builds, or host exposure.
  • The archived patch association depends on search excerpts. No affected host or control was tested here.
  • The check-to-use group has no required shared postcondition; this case does not claim response transfer across the check-to-use group.

Curated archival support: Object-path redirection · method notes · Shared check-to-use group

External references for this claim: "Microsoft fixed Defender flaw RoguePlanet (CVE-2026-50656)." Security Affairs, 9 July 2026. Source for the Windows Server mounting limitation. · "CVE-2026-50656 'RoguePlanet': Microsoft Defender Malware Protection Engine Privilege Escalation to SYSTEM." Threat-Modeling.com, 9 July 2026. Source for engine version 1.1.26060.3008. · "ShieldBreak In-Depth Analysis." CRYPTRON Security GmbH, August 2026. Source for the phase sequence: detected test content served first, payload served on restart hydration, Object Manager links redirected concurrently. Direct retrieval refused by the site's robots policy; search excerpt only.

Provenance: derived

Shared across the three variations

The archive reports no measured requirements

Archive record

The supplied case archive documents it.

Open claim details
Archive record

The historical provenance block records zero measured elements and zero perturbation runs. A documented local correction labels authored source-based requirements asserted and keeps the inferred core and derived products separate. Neither classification establishes measurement or empirical validity.

Claim details and sources
  • Supported here means documented by the supplied local archive. It does not verify the external incident or the archive’s historical test claims.

Curated archival support: Requirement provenance · Declared closed scope

External references for this claim: This claim concerns the supplied archive itself.

Provenance: asserted

Shared across the three variations

The formal bindings carry a pin mismatch

Archive record

The supplied case archive documents it.

Open claim details
Archive record

The archive uses MC-SPEC 0.4.19-draft with corpus hash be8b75136580, while its mc-core 0.4.20.1 ontology pin names 65b2c0f51cf7. The archive marks the bindings provisional. This local presentation preserves that qualification.

Claim details and sources
  • The underlying specification and ontology deposits were not revalidated in this task.
  • The authoring schema and proposed display statuses are not MC normative schemas or enums.

Curated archival support: Formal bindings

External references for this claim: This claim concerns the supplied archive itself.

Provenance: asserted

Applies to Object-path redirection

An unavailable placement is different from an undeployed one

Inferred

The records imply it. The claim details retain its assumptions and supporting records.

Open claim details
Archive record

The archived defensive profile distinguishes a mechanism placement with no available defender-reachable defeating value from an available placement that has not been deployed. An ambient platform condition must not be credited as a defensive action.

Claim details and sources
  • The retained typing describes the archived model; it does not prove that no present-day control can affect the condition.
  • No deployed defensive posture is supplied.

Curated archival support: Object-path redirection · defensive design profile

External references for this claim: This claim concerns the supplied archive itself.

Provenance: derived

Shared across the three variations

Protection on a real host remains unassessed

Unassessed

The case has not evaluated this question.

Open claim details
Archive record

This case contains no deployed posture, detection test or perturbation result. It cannot decide whether a particular host is protected, whether a candidate detector works, or whether the chain is reachable in that environment.

Claim details and sources
  • An enablement edge supplies at least one later clause; paths do not prove full availability or reachability.
  • Frame conditions and parallel-chain interference are outside the archived model’s stated coverage.

Curated archival support: Type-level scope · Requirement provenance · Enablement graph limits

External references for this claim: This claim concerns the supplied archive itself.

Provenance: derived

Related research · outside the curated selection

The execution step (M5) retains both required inputs beyond the execution group (C3) core

Inferred

The records imply it. The claim details retain its assumptions and supporting records.

Open claim details
Archive record

The archived execution step (M5) record requires both attacker-controlled privilegedProcessInput and an attacker-chosen privilegedWriteTarget. The input clause remains a required variation beyond the execution group (C3); neither supplier alone is claimed sufficient.

Claim details and sources
  • The residual execution step (M5) bench test has not been run. The execution group (C3) is separate research context supported by two authored records, below the stated support guideline.

Curated archival support: Execution step (M5) · formal source · Execution step (M5) · required variation · Related execution group (C3)

External references for this claim: This claim concerns the supplied archive itself.

Provenance: derived

Shared across the three variations

Current incident and patch status are outside this snapshot

Unassessed

The case has not evaluated this question.

Open claim details
2 cited sources

The archive is dated 5 September 2026. No external advisory, repository, engine build or patch status was refreshed for this local draft. The case is a dated walkthrough of the model.

Claim details and sources
  • The two primary registry entries were not retrieved in the archive either.
  • Historical “no patch” and affected-build statements are not repeated as current advice.

Curated archival support: Source register · reference 15 · Source register · reference 16

External references for this claim: Microsoft Security Response Center. Security Update Guide entry for CVE-2026-69414. Not retrieved directly; status as reported by references 1 and 9. · CVE Program record, CVE-2026-69414.

Provenance: derived

Source register · 17 dated records

These are historical retrieval labels. Source count and distinct publishers do not establish independent confirmation. Original full texts are not bundled; external destinations require a network connection.

  1. MSNightmare. "ShieldBreak: Windows Defender 0day vulnerability." GitHub repository, August 2026. Repository description only; no exploit content was read, retrieved or reproduced.

    repository-description · repository description only

    Punctuation normalized for display from the archived title; original wording has not been independently checked.

    Original archived title

    MSNightmare. "ShieldBreak " Windows Defender 0day vulnerability." GitHub repository, August 2026. Repository description only; no exploit content was read, retrieved or reproduced.

    Reference S08
  2. "CVE-2026-50656: Microsoft Defender RoguePlanet Zero-Day Enables Local Privilege Escalation on Fully Patched Windows 10 and 11 Systems." Rescana, 18 June 2026.

    secondary-report · search excerpt only

    Punctuation normalized for display from the archived title; original wording has not been independently checked.

    Original archived title

    "CVE-2026-50656: Microsoft Defender RoguePlanet Zero-Day Enables Local Privilege Escalation on Fully Patched Windows 10 " 11 Systems." Rescana, 18 June 2026.

    Reference S10
  3. CVE Program record, CVE-2026-69414.

    primary-registry · not retrieved

    Reference S16

06 / Inspect

The formal record: requirements, mechanism, resulting state

The formal record holds the requirements (the formal name for the conditions the action needs), the mechanism, and the resulting state that the walkthrough text describes. The record also links to its defensive profile (where a defense could act on this action) and to its provenance and compatibility details.

Record set: local-draft-3 · archival source 0.4.0

Content substitution M3

Actual normalized P / M / Q fields · requirement basis: inferred; no perturbation · provenance: asserted

Provenance correction: derived → asserted. Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.

P · Preconditions

  • executionContext ⊒ user
  • hydrationCallbackOwner = adversary
  • scannerAttention = onAdversaryPath
  • contentRevalidation = absent

M · Mechanism

  • serve detected test content on the first hydration, so the engine enters remediation
  • serve different content on the restart hydration that the remediation read triggers

Q · Postconditions

  • privilegedProcessInput = adversaryControlled
Curated formal record slice · JSON
{
  "id": "M3",
  "title": "Swap the file contents during cleanup",
  "formal": {
    "preconditions": [
      "executionContext ⊒ user",
      "hydrationCallbackOwner = adversary",
      "scannerAttention = onAdversaryPath",
      "contentRevalidation = absent"
    ],
    "mechanism": [
      "serve detected test content on the first hydration, so the engine enters remediation",
      "serve different content on the restart hydration that the remediation read triggers"
    ],
    "postconditions": [
      "privilegedProcessInput = adversaryControlled"
    ]
  },
  "variation": {
    "againstCore": "C2",
    "stateVariable": "contentRevalidation",
    "stillRequiredByRecord": true,
    "note": "The served bytes and their revalidation gap are the record’s required variation beyond C2."
  },
  "provenanceKind": "asserted",
  "requirementBasis": "inferred; no perturbation",
  "provenanceCorrection": {
    "originalKind": "derived",
    "correctedKind": "asserted",
    "basis": "Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.",
    "scope": "Authored record requirements only. Cores, graph edges and defensive profile terms retain archived derived status.",
    "executionNote": "This is a local source-level editorial correction. No reasoner or empirical test was run, and historical evidence excerpts retain the original blanket derived label.",
    "modelArtifact": "/artifacts/shieldbreak-model.yaml",
    "changesArtifact": "/artifacts/shieldbreak-model-changes.json"
  },
  "evidenceRefs": [
    "archive-yaml#records/M3"
  ]
}

Object-path redirection M4

Actual normalized P / M / Q fields · requirement basis: inferred; no perturbation · provenance: asserted

Provenance correction: derived → asserted. Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • objectSymlinkCreation ⊒ permittedToUser
  • objectPathRevalidation = absent

M · Mechanism

  • rebind Object Manager symbolic links between the engine's path resolution and its use

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen
Curated formal record slice · JSON
{
  "id": "M4",
  "title": "Redirect the privileged write through an object path",
  "formal": {
    "preconditions": [
      "executionContext ⊒ user",
      "scannerAttention = onAdversaryPath",
      "objectSymlinkCreation ⊒ permittedToUser",
      "objectPathRevalidation = absent"
    ],
    "mechanism": [
      "rebind Object Manager symbolic links between the engine's path resolution and its use"
    ],
    "postconditions": [
      "privilegedWriteTarget = adversaryChosen"
    ]
  },
  "variation": {
    "againstCore": "C2",
    "stateVariable": "objectPathRevalidation",
    "stillRequiredByRecord": true,
    "note": "Object Manager link rebinding and its revalidation gap are the record’s required variation beyond C2."
  },
  "provenanceKind": "asserted",
  "requirementBasis": "inferred; no perturbation",
  "provenanceCorrection": {
    "originalKind": "derived",
    "correctedKind": "asserted",
    "basis": "Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.",
    "scope": "Authored record requirements only. Cores, graph edges and defensive profile terms retain archived derived status.",
    "executionNote": "This is a local source-level editorial correction. No reasoner or empirical test was run, and historical evidence excerpts retain the original blanket derived label.",
    "modelArtifact": "/artifacts/shieldbreak-model.yaml",
    "changesArtifact": "/artifacts/shieldbreak-model-changes.json"
  },
  "evidenceRefs": [
    "archive-yaml#records/M4"
  ]
}

Junction-path redirection N3

Actual normalized P / M / Q fields · requirement basis: inferred; no perturbation · provenance: asserted

Provenance correction: derived → asserted. Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath
  • junctionCreation ⊒ permittedToUser
  • junctionPathRevalidation = absent

M · Mechanism

  • rebind an NTFS junction between the engine's path check and its quarantine write, repeating until the race is won

Q · Postconditions

  • privilegedWriteTarget = adversaryChosen
Curated formal record slice · JSON
{
  "id": "N3",
  "title": "Redirect the privileged write through a folder junction",
  "formal": {
    "preconditions": [
      "executionContext ⊒ user",
      "scannerAttention = onAdversaryPath",
      "junctionCreation ⊒ permittedToUser",
      "junctionPathRevalidation = absent"
    ],
    "mechanism": [
      "rebind an NTFS junction between the engine's path check and its quarantine write, repeating until the race is won"
    ],
    "postconditions": [
      "privilegedWriteTarget = adversaryChosen"
    ]
  },
  "variation": {
    "againstCore": "C2",
    "stateVariable": "junctionPathRevalidation",
    "stillRequiredByRecord": true,
    "note": "NTFS junction rebinding and its revalidation gap are the record’s required variation beyond C2."
  },
  "provenanceKind": "asserted",
  "requirementBasis": "inferred; no perturbation",
  "provenanceCorrection": {
    "originalKind": "derived",
    "correctedKind": "asserted",
    "basis": "Authored source-based requirement declarations; their necessity remains inferred, with no measurement or fresh external validation.",
    "scope": "Authored record requirements only. Cores, graph edges and defensive profile terms retain archived derived status.",
    "executionNote": "This is a local source-level editorial correction. No reasoner or empirical test was run, and historical evidence excerpts retain the original blanket derived label.",
    "modelArtifact": "/artifacts/shieldbreak-model.yaml",
    "changesArtifact": "/artifacts/shieldbreak-model-changes.json"
  },
  "evidenceRefs": [
    "archive-yaml#records/N3"
  ]
}
The shared check-to-use group · C2

Identity: provisional · provenance: derived

Within the archive’s declared closed scope, unasserted facts evaluate as absent. Outside that scope they remain unmodeled, never silently absent or present.

P · Preconditions

  • executionContext ⊒ user
  • scannerAttention = onAdversaryPath

M · Mechanism

  • alter the resolved value between the check and the use

Q · Postconditions

No additional postcondition asserted.

{
  "id": "C2",
  "title": "Change the value between check and privileged use",
  "identityStatus": "provisional",
  "provenanceKind": "derived",
  "observedMembers": [
    "M3",
    "M4",
    "N3"
  ],
  "support": 3,
  "belowSupportFloor": false,
  "membershipNote": "Observed members report the declared population; membership is a structural predicate, not an enumerated definition.",
  "preconditions": [
    "executionContext ⊒ user",
    "scannerAttention = onAdversaryPath"
  ],
  "mechanism": [
    "alter the resolved value between the check and the use"
  ],
  "postconditions": [],
  "plainPreconditions": [
    "The attacker can run code as a normal user.",
    "A privileged security engine is working on an attacker-controlled path."
  ],
  "plainMechanism": "Change the value after a check and before privileged use.",
  "transferRule": "A result transfers only within the supported scope where its placement acts on the invariant; a variation-specific result stays variation-specific.",
  "responseTransfer": "none; C2 has no required shared effect",
  "evidenceRefs": [
    "archive-yaml#invariant_cores/C2",
    "archive-yaml#completeness_claim"
  ],
  "scopeNote": "Within the archive’s declared closed scope, unasserted facts evaluate as absent. Outside that scope they remain unmodeled, never silently absent or present."
}
Versions and bindings
ArtifactVersion / valueMeaning
Local presentation revisionlocal-draft-3authoring wrapper; generated 2026-10-07
Analysis snapshot2026-09-05archived; no external refresh
Record exportCurated formal slice · archival source 0.4.0local-draft-3 curated projection; not a normative release
SpecificationMC-SPEC 0.4.19-draft · be8b75136580archived pin; not revalidated here
Ontologymc-core 0.4.20.1 · spec pin 65b2c0f51cf7provisional binding; pin differs from specification corpus
Fuller public modellocal-draft-3 · archival source 0.4.0source-level correction and documented redactions; not a normative release

The curated formal slice and fuller public model are local-draft-3 projections of archival source 0.4.0, not a new normative release. The fuller model retains ontology typing, state variables, dispositions, defensive profiles and graph fields with documented public redactions. A source-level correction classifies authored requirements as asserted; their necessity remains inferred and unmeasured. The check-to-use group, graph edges and profile terms retain archived derived status. Bindings remain provisional, conceptual ratification unrun and reachability unassessed.

Read the MC semantics →
Method, assumptions and scope limits

Local review draft of the 5 September 2026 type-level analysis, curated to the check-to-use group and content substitution/object-path redirection/junction-path redirection. No external facts were refreshed, no exploit or perturbation was executed, no detector was tested, and no deployed posture was supplied. All modeled requirements are inferred.

The two incidents are attributed in the archive to one researcher. Multiple publishers and distinct URLs are not treated as independent confirmation. Retrieval statuses are historical labels, not the source texts themselves.

Within the archive’s declared closed scope, unasserted facts evaluate as absent. Outside that scope they remain unmodeled, never silently absent or present.

MC-SPEC §1.5: the conceptual ratification suite has not been run against the bound text. Local presentation and structural checks do not replace it.

Assumptions

  • Requirements, placement availability and cores are reconstructed at type level from the supplied archive.
  • Historical reported patch association motivates a scope comparison; it is not treated as a current threat fact.
  • No undisclosed binary, engine internals, platform variance or post-escalation behavior was examined.
  • Within the archive’s declared closed scope, unasserted facts evaluate as absent. Outside that scope they remain unmodeled, never silently absent or present.

Limitations

  • External source texts are not included in the supplied package; archived citations and retrieval labels are preserved without fresh retrieval.
  • The specification/ontology binding remains provisional because the recorded corpus pins differ.
  • Graph paths do not establish full precondition satisfaction, reachability or protection.
  • The archive leaves frame conditions and parallel-chain interference unspecified.
  • Corpus-local extensions and source wrappers do not become normative MC terms.
  • MC-SPEC §1.5: the conceptual ratification suite has not been run against the bound text. Local presentation and structural checks do not replace it.