Aiceberg Risk Signals Library

Aiceberg Risk Signals Library is a collection of AI threat detection and risk intelligence tools designed to secure generative AI model deployments. The library provides detection capabilities for sensitive data exposure including PII (social security numbers, addresses, emails), PHI (medical history, treatment information, insurance details), and PCI data (credit card numbers, expiration dates, CVV).

The product identifies security threats such as secrets (passwords, API keys, cryptographic keys), toxicity, illegal content, and code vulnerabilities including prompt injection, jailbreaking, prompt leaking, and role impersonation. It includes input and output manipulation detection to prevent instruction override, direct command injection, and prompt leaking.

The library offers content control features including blocklists for restricting specific words or topics, system instruction classification, relevance checking, and intent understanding. It provides specialized detection for code presence and code requests, text-to-SQL translation accuracy, and instruction-to-action alignment.

Additional capabilities include goal alignment verification, data loss protection against defined ground truths, and intent-to-instruct validation to minimize misalignment and unintended consequences. The library is continuously expanding and supports enterprise compliance requirements for AI deployments.