Guardrails
Security controls aligned with OWASP LLM TOP 10 and MITRE ATLAS
LLM01: Prompt Injection Protection
Blocks malicious prompts that try to manipulate the AI's behavior or override system instructions.
LLM02: Sensitive Information Disclosure
Automatically detects and removes personal information, credentials, and confidential data from AI responses.
LLM03: Supply Chain Security
Verifies AI models come from trusted sources and scans dependencies for vulnerabilities.
LLM04: Data and Model Poisoning Prevention
Monitors for corrupted training data or model tampering that could affect AI accuracy or security.
LLM05: Improper Output Handling
Validates and sanitizes all AI outputs to prevent security exploits like code injection or XSS attacks.
LLM06: Excessive Agency Controls
Limits AI actions to approved scope and requires human oversight for critical decisions.
LLM07: System Prompt Leakage Prevention
Prevents internal system instructions from being exposed in AI responses.
LLM08: Vector and Embedding Security
Protects knowledge bases and vector stores from unauthorized access or data exposure.
LLM09: Misinformation Prevention
Detects and flags false or misleading information in AI-generated content.
LLM10: Unbounded Consumption Protection
Enforces rate limits and usage quotas to prevent resource exhaustion or denial-of-service attacks.
Language Models
AI language models used and their configurations
| Model | Provider | Purpose | Model Card |
|---|---|---|---|
| Claude Opus 4.5 | Anthropic | Safety testing and Red teaming | View |
Model Training
Information about how customer data is used for model training
Customer Data Usage for Training
Customer data is not used to train AI models. All training is done using publicly available or proprietary datasets.
Data Access and Tools
Third-party integrations and data sources available to AI agents
No data access tools configured yet.
Safety Reports
Security audits, compliance certifications, and safety assessments
No safety reports available yet.