AI SECURITY CASES × FRAMEWORKS
Databricks DASF 3.0 · Categories and AI Security Cases
Explore AI security cases, sources, and defensive analysis across 73 populated Databricks DASF 3.0 categories.
A structured risk assessment framework covering 97 AI security risks across data operations, model lifecycle, deployment, platform, and agentic AI, published by Databricks.
- 1.1 · Insufficient access controls 9 cases
- 1.2 · Missing data classification 0 cases
- 1.3 · Poor data quality 2 cases
- 1.4 · Ineffective storage and encryption 2 cases
- 1.5 · Lack of data versioning 0 cases
- 1.6 · Insufficient data lineage 0 cases
- 1.7 · Lack of data trustworthiness 7 cases
- 1.8 · Legality of data 2 cases
- 1.9 · Stale data 0 cases
- 1.10 · Lack of data access logs 1 cases
- 1.11 · Compromised 3rd-party datasets 3 cases
- 2.1 · Preprocessing integrity 1 cases
- 2.2 · Feature manipulation 0 cases
- 2.3 · Raw data criteria 1 cases
- 2.4 · Adversarial partitions 0 cases
- 3.1 · Data poisoning 5 cases
- 3.2 · Ineffective storage and encryption 1 cases
- 3.3 · Label flipping 0 cases
- 4.1 · Lack of traceability and transparency of model assets 1 cases
- 4.2 · Lack of end-to-end ML lifecycle 0 cases
- 5.1 · Lack of tracking and reproducibility of experiments 0 cases
- 5.2 · Model drift 0 cases
- 5.3 · Hyperparameters stealing 0 cases
- 5.4 · Malicious libraries 12 cases
- 6.1 · Evaluation data poisoning 0 cases
- 6.2 · Insufficient evaluation data 0 cases
- 6.3 · Lack of Interpretability and Explainability 0 cases
- 7.1 · Backdoor machine learning / Trojaned model 9 cases
- 7.2 · Model assets leak 8 cases
- 7.3 · ML Supply chain vulnerabilities 10 cases
- 7.4 · Source code control attack 7 cases
- 8.1 · Model attribution 0 cases
- 8.2 · Model theft 11 cases
- 8.3 · Model lifecycle without HITL (human-in-the-loop) 1 cases
- 8.4 · Model inversion 0 cases
- 9.1 · Prompt inject 27 cases
- 9.2 · Model inversion 5 cases
- 9.3 · Model breakout 20 cases
- 9.4 · Looped input 3 cases
- 9.5 · Infer training data membership 0 cases
- 9.6 · Discover ML model ontology 1 cases
- 9.7 · Denial of Service (DoS) 1 cases
- 9.8 · LLM hallucinations 2 cases
- 9.9 · Input Resource Control 43 cases
- 9.10 · Accidental exposure of unauthorized data to models 3 cases
- 9.11 · Model Inference API Access 19 cases
- 9.12 · LLM Jailbreak 6 cases
- 9.13 · Excessive agency 61 cases
- 10.1 · Lack of audit and monitoring inference quality 0 cases
- 10.2 · Output manipulation 2 cases
- 10.3 · Discover ML model ontology 1 cases
- 10.4 · Discover ML model family 3 cases
- 10.5 · Black-box attacks 6 cases
- 10.6 · Sensitive data output from a model 39 cases
- 11.1 · Lack of MLOps — repeatable enforced standards 0 cases
- 12.1 · Lack of vulnerability management 1 cases
- 12.2 · Lack of penetration testing, red teaming and bug bounty 0 cases
- 12.3 · Lack of incident response 3 cases
- 12.4 · Unauthorized privileged access 1 cases
- 12.5 · Poor security in the software development lifecycle 15 cases
- 12.6 · Lack of compliance 0 cases
- 12.7 · Initial Access 51 cases
- 13.1 · Memory Poisoning 15 cases
- 13.2 · Tool Misuse 57 cases
- 13.3 · Privilege Compromise 25 cases
- 13.4 · Resource Overload 1 cases
- 13.5 · Cascading Hallucination Attacks 1 cases
- 13.6 · Intent Breaking & Goal Manipulation 52 cases
- 13.7 · Misaligned & Deceptive Behaviors 16 cases
- 13.8 · Repudiation & Untraceability 3 cases
- 13.9 · Identity Spoofing & Impersonation 4 cases
- 13.10 · Overwhelming Human in the Loop 0 cases
- 13.11 · Unexpected RCE and Code Attacks 49 cases
- 13.12 · Agent Communication Poisoning 4 cases
- 13.13 · Rogue Agents in Multi-Agent Systems 5 cases
- 13.14 · Human Attacks on Multi-Agent Systems 5 cases
- 13.15 · Human Manipulation 9 cases
- 13.16 · Prompt Injection 4 cases
- 13.17 · Confused Deputy 9 cases
- 13.18 · Tool Poisoning 3 cases
- 13.19 · Credential and Token Exposure 31 cases
- 13.20 · Insecure Server Configuration 5 cases
- 13.21 · Supply Chain Attacks 9 cases
- 13.22 · Excessive Permissions and Scope Creep 3 cases
- 13.23 · Data Exfiltration 7 cases
- 13.24 · Context Spoofing and Manipulation 49 cases
- 13.25 · Insecure Communication 0 cases
- 13.26 · Malicious Server Connection 8 cases
- 13.27 · Insecure Credential Storage 1 cases
- 13.28 · UI/UX Deception 2 cases
- 13.29 · Insufficient Server Validation 6 cases
- 13.30 · Client-Side Data Leakage 3 cases
- 13.31 · Excessive Permission Granting 1 cases
- 13.32 · Client-Side Code Execution 4 cases
- 13.33 · Insecure Communication Handling 0 cases
- 13.34 · Session and State Management Failures 3 cases
- 13.35 · Update and Patch Management 4 cases
An empty category does not establish safety. A category relation does not prove a defense is deployed or effective.