GPT-4o
Auditor
BETAEvaluate the security of your website Chatbots, models, and code under several compliance frameworks.
Website
Evaluate security of chatbot AI models on websites
API
Evaluate your local or server-deployed AI models via API
GitHub
Upload GitHub code to evaluate code security
Overview
Frameworks
Attacks
News
Agents-Risks
Data
AI Security Risks & Threats
Understanding potential dangers in AI systems
LLM Vulnerabilities
Prompt injection attacks, data poisoning, and model manipulation can expose sensitive information or generate harmful content.
View DetailsAgent Security
AI agents with excessive permissions may leak credentials, execute unauthorized commands, or cause system compromises.
View DetailsMCP Tool Risks
Model Context Protocol tools can be exploited to access banking systems, steal credentials, or perform unauthorized transactions.
View DetailsCode Vulnerabilities
Traditional security flaws in applications that integrate AI systems, creating attack vectors through conventional weaknesses.
View DetailsPrivacy Exposure
Personal data leakage through training data memorization, prompt engineering, or inadequate output filtering mechanisms.
View DetailsHarmful Content
Generation of inappropriate material including violence, terrorism, pornography, or other content that violates safety policies.
View DetailsGenAI RMF
NIST's Generative AI Risk Management Framework provides guidelines for identifying, assessing, and mitigating risks in generative AI systems.
View DetailsNIST CSF
Cybersecurity Framework that provides guidelines, standards, and best practices to manage cybersecurity risks including AI security concerns.
View DetailsOWASP Top 10
Open Web Application Security Project's Top 10 for Large Language Model Applications - identifying the most critical security risks.
View DetailsATLAS Matrix
MITRE's Adversarial Threat Landscape for Artificial-Intelligence Systems - a comprehensive framework for AI threat modeling.
View DetailsPrompt Exfiltration
Advanced techniques to extract system prompts and internal instructions from AI models through carefully crafted input sequences.
View DetailsJailbreaking
Sophisticated methods to bypass safety guardrails and content filters to make AI models generate prohibited or harmful content.
View DetailsSelf-Modification
Attacks that manipulate AI models to modify their own behavior, instructions, or operational parameters during runtime.
View DetailsCovert Channel
Hidden communication methods through AI model outputs that can exfiltrate data or transmit unauthorized information.
View Details
.png)
Agent Privilege Escalation
Autonomous agents gaining unauthorized access levels, potentially compromising entire systems through elevated permissions.
View DetailsInter-Agent Communication
Security risks in agent-to-agent communication including message interception, manipulation, and unauthorized coordination.
View DetailsCredential Exposure
Agents inadvertently exposing API keys, passwords, or sensitive authentication tokens through logs or outputs.
View DetailsRecursive Execution
Agents creating infinite loops or recursive operations that can cause denial of service or resource exhaustion attacks.
View DetailsTraining Data Poisoning
Malicious manipulation of training datasets to introduce backdoors, bias, or vulnerabilities into AI models.
View DetailsData Leakage
Unintended exposure of sensitive training data through model outputs, inference attacks, or membership inference.
View DetailsModel Extraction
Adversaries reverse-engineering or stealing proprietary models through carefully crafted queries and analysis.
View DetailsAdversarial Examples
Specially crafted inputs designed to fool AI models into making incorrect predictions or classifications.
View Details