Versatile AI Risk Assessment

Source directoryBerryville Institute of Machine Learning (BIML)v1.0 (2024-01-24)

An Architectural Risk Analysis of Large Language Models: Applied Machine Learning Security

“An Architectural Risk Analysis of Large Language Models: Applied Machine Learning Security”. Publisher: Berryville Institute of Machine Learning (BIML). Version used: v1.0 (2024-01-24). The Versatile AI Risk Assessment threat catalogue backs 25 of its 52 threats with this document, at 20 verified locations.

As of: July 2026 · Catalogue version 2026.07.17.3

Document details

Publisher
Berryville Institute of Machine Learning (BIML)
Version used
v1.0 (2024-01-24)
Year
2024
Licence
CC BY-SA 4.0 International
Framework
BIML

Open the original document at the publisher External link; the publisher’s version always takes precedence.

References in the catalogue (20)

Every row states the clause, its title, the exact location in the document and the threats that refer to it.

Clauses of this document with the threats that reference them
Clause Title and location Referencing threats
BIML-LLM inference:1 Prompt Manipulation (Aka Prompt Injection) PDF p. 18, [inference:1:prompt manipulation (aka prompt injection)] Jailbreaks
BIML-LLM inference:10 User Risk PDF p. 19, [inference:10:user risk] Sensitive Information DisclosureApplication VulnerabilitiesShadow AI (Unsanctioned AI Service Use)
BIML-LLM inference:3 Wrongness PDF p. 18, [inference:3:wrongness] Factual Inconsistencies (Hallucinations)
BIML-LLM inference:8 Unstructured Output PDF p. 19, [inference:8:unstructured output] Insecure Output Handling
BIML-LLM inference:9 Hosting PDF p. 19, [inference:9:hosting] Supply Chain – InfrastructureApplication VulnerabilitiesApplication Denial of Service
BIML-LLM input:2 Prompt Injection PDF p. 16, [input:2:prompt injection] Prompt Injection – DirectPrompt Injection – Indirect
BIML-LLM input:3 Open to the Public PDF p. 16, [input:3:open to the public] JailbreaksModel Reconnaissance
BIML-LLM input:5 Sponge Input PDF p. 17, [input:5:sponge input] Model Denial of Service
BIML-LLM LLMtop10:2 Data Debt PDF p. 12, [LLMtop10:2:data debt] Supply Chain – Datasets
BIML-LLM LLMtop10:5 Prompt Manipulation PDF p. 13, [LLMtop10:5:prompt manipulation] Prompt Injection – DirectPrompt Injection – IndirectJailbreaks
BIML-LLM LLMtop10:6 Poison in the Data PDF p. 13, [LLMtop10:6:poison in the data] Supply Chain – DatasetsTraining Data PoisoningTargeted Poisoning / Label Poisoning
BIML-LLM LLMtop10:9 Model Trustworthiness PDF p. 13, [LLMtop10:9:model trustworthiness] OverrelianceFactual Inconsistencies (Hallucinations)
BIML-LLM model:4 Trojan PDF p. 17, [model:4:Trojan] Supply Chain – ModelsBackdoor ML ModelSleepy Agent (Time/Event-Triggered Hidden Instructions)
BIML-LLM model:6 Training Set and Prompt Reveal PDF p. 18, [model:6:training set and prompt reveal] Meta Prompt ExtractionPrivacy Attacks
BIML-LLM model:9 Modality PDF p. 18, [model:9:modality] Multimodal Attacks
BIML-LLM output:11 Black Box Discrimination PDF p. 20, [output:11:black box discrimination] Hate Speech and Discrimination
BIML-LLM output:12 Overconfidence PDF p. 20, [output:12:overconfidence] Overreliance
BIML-LLM raw:10 Query Data PDF p. 15, [raw:10:query data] Prompt Injection – IndirectFactual Inconsistencies (Hallucinations)RAG-Specific Attacks (Document Poisoning)
BIML-LLM raw:5 Data Confidentiality PDF p. 15, [raw:5:data confidentiality] Privacy AttacksSensitive Information Disclosure
BIML-LLM raw:9 Time PDF p. 15, [raw:9:time] Model Drift & Degradation

This page does not reproduce the text of the standards. It states the identifier, title and location of the clause; the wording itself is in the original document. The mappings are taxonomic and not evidence of compliance.

Threats referencing this document (25)

Grouped by topic. Every entry leads to the full threat page.

Application and Integration Security

Attacks on the Running Model and Service

Harmful Content

Model and Training Data Manipulation

Privacy and Data Leakage

Prompt Attacks and Guardrail Evasion

Reliability and Responsible Use

Supply Chain and Provenance

More documents from the same publisher

All documents by Berryville Institute of Machine Learning (BIML) cited in the catalogue.

An Architectural Risk Analysis of Machine Learning Systems: Toward More Secure Machine Learning v1.0 (2020-01-13) · 2020

From the reference to the assessment

The full catalogue states the mitigations, the possible impact and every verified location for each threat. The live demo runs locally in your browser, with no sign-up.

Go to the threat catalogue Start the live demo

Cite this page

For reports, policies or internal documents; the link leads directly to this source page.

“An Architectural Risk Analysis of Large Language Models: Applied Machine Learning Security” (Berryville Institute of Machine Learning (BIML)). References in the AI threat catalogue, Versatile AI Risk Assessment, as of July 2026.
https://www.versatile-ai-risk-assessment.com/en/wissensbasis/sources/biml-architectural-risk-analysis-large-language-models/

← Back to the source directory