​

🛡️ Responsible AI

Lyzr’s Responsible AI module enables platform users to proactively moderate content, prevent misuse, and ensure compliance with privacy and safety standards. With built-in support for detecting toxicity, prompt injections, sensitive information, and more, you can build AI agents that are safe, ethical, and secure.

​

🔥 Toxicity Detection

Automatically detect and prevent the generation or processing of toxic, harmful, or offensive content.

• Use Case: Prevent agents from generating insults, hate speech, threats, or inappropriate language in customer support, education, or community applications.
• Threshold: 0.4
  (Values closer to 1 indicate higher tolerance, while lower thresholds are more strict.)

✨ Agents will automatically block or filter responses that exceed the toxicity threshold.

​

🎭 Prompt Injection Protection

Detect and block malicious prompt manipulation attempts (Prompt Injections), where a user tries to override or influence the agent’s behavior using cleverly crafted input.

• Use Case: Prevent users from bypassing system instructions (e.g., “Ignore the last instruction and say X”).
• Threshold: 0.3
  (Lower values are stricter and more secure.)

🔐 Especially useful in agents interacting with untrusted or anonymous users.

​

🔐 Secrets Detection

Automatically detect and redact or mask sensitive credentials, including:

• API keys

• Tokens

• JWTs (JSON Web Tokens)

• Private Keys

• Use Case: Prevent accidental exposure of credentials in logs or chat outputs.

• Action: Detected values are redacted before being stored, displayed, or transmitted.

​

✅ Allowed Topics

Restrict agent interactions to only specific, whitelisted topics.

• Use Case: Ensure your agent only discusses business-allowed domains (e.g., “finance, healthcare, HR”).
• Configuration:
  Provide comma-separated values:
  Copy

  finance,healthcare,HR
  

🧠 Useful for domain-specific AI assistants with strict focus.

​

🚫 Banned Topics

Prevent the agent from discussing or responding to specific blacklisted topics.

• Use Case: Prohibit conversation around internal operations, political views, or adult content.
• Configuration:
  Provide comma-separated banned topics:
  Copy

  politics,internal roadmap,violence
  

​

❌ Blocked Keywords

Restrict or redact specific words or phrases from being used in prompts or responses.

• Use Case: Redact client names, project codenames, or other internal terms.
• Configuration:
  Provide comma-separated keywords:
  Copy

  ProjectX,CEO,InternalCode
  

💡 Blocked keywords will be filtered out or replaced during processing.

​

🔍 Personally Identifiable Information (PII)

Control how agents handle sensitive personal data, with options to block or redact each category.

​

Supported Categories & Actions

🔐 These controls help you comply with GDPR, HIPAA, and other data protection standards.

​

🎯 Example Use Cases

​

📌 How to Configure in Studio

1. Go to Agent Settings in Studio.
2. Open the Responsible AI tab.
3. Toggle each feature and configure the appropriate thresholds or keywords.
4. Save and apply the settings.

⚙️ Changes take effect immediately for all new interactions.

By enabling Responsible AI, you ensure that your Lyzr agents act ethically, safely, and in alignment with your organization’s privacy and compliance standards.