Risk Signals Library

Discerning special entities such as social security numbers, date of birth, addresses, emails, etc.

Discerning special entities such as medical history, treatment information, insurance details, etc.

Discerning special entities such as credit card numbers, exp date, CVV

Passwords, API Keys, crypto keys, etc.

Identifying and mitigating harmful or inappropriate language

Preventing content that may violate laws

Restrict specific words, phrases, or topics from being processed or generated by the AI

Ensures that the model's responses and actions are in direct correspondence with the instructions provided by users

Ensures the content generated is pertinent to the context of the interaction

Understanding and aligning with the user's purpose

Discerning special entities such as medical history, treatment information, insurance details, etc.

Ensures that executable content is only included when explicitly requested by the user

Tactics like prompt injection, instruction override, or direct command injection are identified and neutralized

Stops the leaking of prompts that could reveal sensitive information or internal system data

Ensures AI’s actions remain aligned with intended objectives and user directives.

Syntactic instructions and semantic based attacks such as prompt injection, jailbreaking, prompt leaking or role impersonation

Ensures accuracy and relevance in tasks that require precise language-to-code translation

Harmonizes the user's stated objectives and intents with the actual actions performed by the AI

We analyze the content against the defined data loss ground truth and alert or enforce policy

Ensuring that AI correctly interprets and follows the intended instruction of a prompt while minimizing the risk of misalignment, harmful outputs, or unintended consequences