OWASP TOP 10 LLM VULNERABILITIES
Home
OWASP
LLM 01
Prompt Injection
Attackers manipulate input prompts to compromise model outputs and behavior.
How AHAGuardians mitigate this risk?
Robust Prompt Sanitization
Filters and neutralizes malicious prompt injections.
Controlled Prompt Templates
Enforces approved prompt structures, preventing free-form input.
Dynamic Prompt Parameterization
Safely inserts user data into prompts, minimizing injection risk.
LLM 02
Sensitive Information Disclosure
Unintended exposure of sensitive information during model operation.
How AHAGuardians mitigate this risk?
Input/Output Content Filtering
Detects and redacts PII, preventing exposure.
Data Loss Prevention (DLP) Integration
Enforces data security policies.
Contextual Data Masking
Dynamically masks sensitive data within outputs.
LLM 03
Supply Chain
Vulnerabilities arising from compromised model development and deployment elements.
How AHAGuardians mitigate this risk?
Model Provenance Tracking
Tracks model components for quick identification of compromises.
Secure Model Deployment Pipelines
Ensures only authorized models are deployed.
Dependency Vulnerability Scanning
Scans dependencies for known vulnerabilities.
LLM 04
Data and Model Poisoning
Introducing malicious data or poisoning the model to manipulate its behavior.
How AHAGuardians mitigate this risk?
Training Data Validation
Validates data for anomalies and poisoning
Model Performance Monitoring
Monitors for deviations indicating poisoning.
Differential Privacy Techniques
Mitigates poisoned data impact during training.
LLM 05
Improper Output Handling
Flaws in managing and safeguarding generated content, risking unintended consequences.
How AHAGuardians mitigate this risk?
Output Content Moderation
Filters harmful or inappropriate outputs.
Output Validation and Sanitization
Validates and sanitizes outputs.
Human-in-the-Loop Review
Enables human review for critical applications.
LLM 06
Excessive Agency
Overly permissive model behaviors that may lead to undesired outcomes.
How AHAGuardians mitigate this risk?
Controlled Model Access
Defines fine-grained access control policies.
Workflow Automation with Human Oversight
Requires human approval for critical actions.
Output Confinement
Restricts the scope of LLM outputs.
LLM 07
System Prompt Leakage
Leakage of internal prompts exposing the operational framework of the LLM.
How AHAGuardians mitigate this risk?
Secure Prompt Storage
Securely stores and encrypts system prompts.
Prompt Access Control
Implements strict access control for prompts.
Prompt Versioning and Auditing
Maintains version history and logs access.
LLM 08
Vector and Embedding Weaknesses
Weaknesses in vector storage and embedding representations that may be exploited.
How AHAGuardians mitigate this risk?
Secure Vector Database Integration
Integrates with secure vector databases.
Embedding Analysis and Monitoring
Analyzes and monitors embeddings for anomalies.
Adversarial Embedding Detection
Detects manipulative embeddings.
LLM 09
Misinformation
LLMs inadvertently generating or propagating misinformation.
How AHAGuardians mitigate this risk?
Bias Detection and Mitigation
Detects and mitigates biases in outputs.
Source Attribution
Provides source attribution for outputs.
Fact Verification Integratio
Integrates with fact-checking services.
LLM 10
Unbounded Consumption
Uncontrolled resource consumption by LLMs, causing service disruptions.
How AHAGuardians mitigate this risk?
Resource Quotas and Limits
Defines resource quotas for LLM usage.
Usage Monitoring and Alerting
Monitors usage and alerts on unusual activity.
Cost Management
Tracks spending and optimizes resource allocation.