CVE-2025-6211 — MEDIUM (CVSS 6.5) AI Security Vulnerability

CISO Take

LlamaIndex's DocugamiReader silently loses document chunks when distinct sections share identical text due to MD5 hash collisions, causing AI responses to be based on incomplete or wrong context with no error surfaced. This is a critical integrity risk for any RAG or document Q&A pipeline handling Docugami-formatted documents—especially in legal, compliance, or audit workflows where missed clauses go undetected. Patch to llama-index-readers-docugami >= 0.3.1 and llama-index >= 0.12.41, then re-index all previously processed document stores.

Risk Assessment

The medium CVSS score (6.5) understates operational risk for AI pipelines. Silent chunk loss means affected systems hallucinate or omit critical content with no visible error or alert—a severe integrity failure in regulated industries. EPSS of 0.00067 indicates no observed active exploitation. The vulnerability triggers passively under normal usage: repeated clauses (boilerplate, standard terms) common in business and legal documents are sufficient to cause collisions without adversarial input. Risk is highest for organizations using LlamaIndex with Docugami in legal review, compliance evidence, or regulated document workflows.

Affected Systems

Package	Ecosystem	Vulnerable Range	Patched
llama-index	pip	< 0.12.41	`0.12.41`
49.1K 229 dependents Pushed 8d ago 87% patched ~50d to patch Full package profile →
llama-index-readers-docugami	pip	< 0.3.1	`0.3.1`
49.1K 229 dependents Pushed 8d ago 87% patched ~50d to patch Full package profile →

Severity & Risk

CVSS 3.1

6.5 / 10

EPSS

0.3%

chance of exploitation in 30 days

Higher than 53% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

Composite signal derived from CISA KEV, CISA SSVC, EPSS, trickest/cve, and Nuclei templates.

Attack Surface

AV Network

AC Low

PR None

UI None

S Unchanged

C None

I Low

A Low

Recommended Action

5 steps

Patch immediately: upgrade llama-index-readers-docugami to >= 0.3.1 and llama-index to >= 0.12.41.
Re-index: all documents processed with affected versions must be re-ingested after patching to restore chunk integrity.
Audit outputs: cross-check AI responses against source documents for any critical legal, compliance, or contractual content generated before patching.
Detect past exposure: compare chunk counts before and after re-indexing—significant drops confirm prior collisions.
Workaround if patching is delayed: prepend positional or structural metadata to chunk text before hashing to enforce uniqueness at the application layer.

CISA SSVC Assessment

Decision Track*

Exploitation poc

Automatable Yes

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

Classification

Supply Chain Data Leakage Framework RAG AML.T0010.001 - AI Software AML.T0031 - Erode AI Model Integrity AML.T0059 - Erode Dataset Integrity AML.T0070 - RAG Poisoning

Compliance Impact

This CVE is relevant to:

EU AI Act

Article 10 - Data and data governance

ISO 42001

8.4 - Data for AI systems

NIST AI RMF

MAP 2.2 - Data quality and integrity risks to AI

OWASP LLM Top 10

LLM03:2025 - Supply Chain LLM08:2025 - Vector and Embedding Weaknesses

Frequently Asked Questions

What is CVE-2025-6211?

LlamaIndex's DocugamiReader silently loses document chunks when distinct sections share identical text due to MD5 hash collisions, causing AI responses to be based on incomplete or wrong context with no error surfaced. This is a critical integrity risk for any RAG or document Q&A pipeline handling Docugami-formatted documents—especially in legal, compliance, or audit workflows where missed clauses go undetected. Patch to llama-index-readers-docugami >= 0.3.1 and llama-index >= 0.12.41, then re-index all previously processed document stores.

Is CVE-2025-6211 actively exploited?

No confirmed active exploitation of CVE-2025-6211 has been reported, but organizations should still patch proactively.

How to fix CVE-2025-6211?

1. Patch immediately: upgrade llama-index-readers-docugami to >= 0.3.1 and llama-index to >= 0.12.41. 2. Re-index: all documents processed with affected versions must be re-ingested after patching to restore chunk integrity. 3. Audit outputs: cross-check AI responses against source documents for any critical legal, compliance, or contractual content generated before patching. 4. Detect past exposure: compare chunk counts before and after re-indexing—significant drops confirm prior collisions. 5. Workaround if patching is delayed: prepend positional or structural metadata to chunk text before hashing to enforce uniqueness at the application layer.

What systems are affected by CVE-2025-6211?

This vulnerability affects the following AI/ML architecture patterns: RAG pipelines, document processing pipelines, legal document analysis, compliance document workflows.

What is the CVSS score for CVE-2025-6211?

CVE-2025-6211 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.30%.

Technical Details

NVD Description

A vulnerability in the DocugamiReader class of the run-llama/llama_index repository, up to but excluding version 0.12.41, involves the use of MD5 hashing to generate IDs for document chunks. This approach leads to hash collisions when structurally distinct chunks contain identical text, resulting in one chunk overwriting another. This can cause loss of semantically or legally important document content, breakage of parent-child chunk hierarchies, and inaccurate or hallucinated responses in AI outputs. The issue is resolved in version 0.3.1.

Exploitation Scenario

An attacker with document upload access crafts a contract or policy document with identical clause text in structurally distinct positions—a common pattern in real legal documents (boilerplate, repeated definitions). When DocugamiReader processes the document, the MD5 hash collision causes the later chunk to silently overwrite the earlier one, deleting an obligation, liability clause, or exclusion from the index. The AI assistant subsequently confirms compliance or contract terms based on the corrupted index, with the dropped clause entirely absent from its context. In a compliance evidence workflow, this could result in an audit gap going undetected until challenged externally. No attacker is required for passive exploitation—natural repetition in real business documents triggers collisions without any adversarial input.