OBS-2026-000223: CVE-2026-111392
1. Observe What changed? โ
2. Understand Why does it matter? โ
3. Track How does risk evolve? โ
4. Predict What happens next? โ
5. Decide What to do? โ
6. Verify What happened? โ
7. Remember Persistent memory
IMMUTABLE CANONICAL OBSERVATION
OBS-2026-000223
Target Entity: CVE-2026-111392 โ SGLang Multimodal RadixAttention KV-Cache Poisoning & Cross-Tenant Image Disclosure via Caller-Declared mm_hashes
MODE: LIVE
Methodology: HTS-3.1
Observed UTC:
2026-10-17T00:00:00.000ZEvent Date:
2026-10-17Epistemic State:
โ KNOWN
โ KNOWN
Cryptographic Digest:
sha256-5d02c975804c9f6โฆ1. Verified Evidence & Observed State
Section titled โ1. Verified Evidence & Observed StateโThis canonical observation confirms the following factual event on the threat surface:
Hermes Threat Score (HTS)
91 / 100
CRITICAL
CVSS v3.1
7.5
EPSS (30-Day Exploitation)
31.8%
Percentile: 92.4th
Weaponization & CISA KEV
๐ก๏ธ Not in KEV
Exploit Maturity: poc
2. Decomposed Epistemic Confidence Vector
Section titled โ2. Decomposed Epistemic Confidence VectorโHermes rejects false certainty and explicitly exposes its decomposed confidence vector:
Evidence Integrity94%
Model Stability98%
Data Completeness95%
Actionable Decisiveness85%
3. Primary Sources & Provenance Chain
Section titled โ3. Primary Sources & Provenance ChainโObservation OBS-2026-000223 is strictly anchored to verified records from the following primary entities:
- VulnCheck Advisory โ SGLang Through 0.5.21 KV-Cache Poisoning via mm_hashes in /generate (CVE-2026-111392)
- SGLang GitHub Issue #44641 โ Unauthenticated KV cache poisoning and cross-request image disclosure via caller-declared mm_hashes
- SGLang GitHub Pull Request #44653 โ Gate caller-supplied mm_hashes behind โtrust-caller-mm-hashes
- NIST National Vulnerability Database โ CVE-2026-111392
Causal Intelligence Lineage
Section titled โCausal Intelligence LineageโPrimary Source โ OBS-2026-000223 โ CVE-2026-111392 โ Risk Trajectory โ Action Directive
4. Target Entity & Correlated Intelligence
Section titled โ4. Target Entity & Correlated Intelligenceโ- Primary Vulnerability Dossier: CVE-2026-111392
- Impacted Software Product :
SGLang Structured Generation & LLM Inference Engine(sgl-project ecosystem) - Dynamic Risk Trajectories: View risk trajectory analysis
- Decision Engine: View prescriptive remediation directives
5. Observation Sequence & Chain of Custody
Section titled โ5. Observation Sequence & Chain of CustodyโChronological lineage for entity CVE-2026-111392:
- Preceding Observation: None (Initial Baseline)
- Succeeding Observation:
OBS-2026-000224
| Observation ID | Date | Mode | Score | Observed Fact |
|---|---|---|---|---|
๐ OBS-2026-000223 | 2026-10-17 | LIVE | 91 | Initial baseline observation created upon public disclosure and ingestion. |
OBS-2026-000224 | 2026-10-18 | LIVE | 91 | State change observed: threat score adjusted by +0 points; CISA KEV or weaponized exploit verified. |
6. Reproducibility & Cryptographic Verification
Section titled โ6. Reproducibility & Cryptographic Verificationโ(Section 77 Mandate: โObservations must be reproducibleโ)
Fetch and verify this observation record directly from the machine-readable API:
# 1. Fetch canonical observation recordcurl -sSL https://hermes-codex.vercel.app/api/observations/OBS-2026-000223.json
# 2. Verify cryptographic SHA-256 integritycurl -sSL https://hermes-codex.vercel.app/api/observations/OBS-2026-000223.json | sha256sumOASIS STIX 2.1 Mapping (observed-data)
Section titled โOASIS STIX 2.1 Mapping (observed-data)โ{ "type": "observed-data", "spec_version": "2.1", "id": "observed-data--8c2a04ad-98b9-63f7-6bd2-72482387cdd5", "created_by_ref": "identity--hermes-codex-threat-observatory", "created": "2026-10-17T00:00:00.000Z", "modified": "2026-10-17T00:00:00.000Z", "first_observed": "2026-10-17T00:00:00.000Z", "last_observed": "2026-10-17T00:00:00.000Z", "number_observed": 1, "external_references": [ { "source_name": "hermes-codex", "external_id": "OBS-2026-000223", "url": "https://hermes-codex.vercel.app/observations/obs-2026-000223/" } ], "object_marking_refs": [ "marking-definition--tlp-clear" ], "custom_properties": { "x_hermes_threat_score": 91, "x_hermes_severity": "CRITICAL", "x_hermes_epistemic_state": "KNOWN", "x_hermes_methodology": "HTS-3.1", "x_hermes_integrity_hash": "sha256-5d02c975804c9f61dd0dec62718bf8e47b81cc407385f183d24637c214e6f132" }}