Transparency

Our AI assistants differ in hosting location, data retention, and purpose. We disclose these details clearly so you can make an informed choice:

• Indefinite retention means we store conversation data unless you request deletion. EU-based assistants automatically delete conversations after 90 days, and “Temporary Chats” are purged after 30 days.

See further detail on the original page.

Explainability

Explainability means showing how and why each AI assistant arrives at its recommendations, without overwhelming you with technical jargon.

1. Reasoning Summaries
   • We present short “Reasoning Summaries” for our AI outputs, highlighting the primary factors or references (e.g., “According to ISO 27001 control 6.1.2…”).
   • These summaries explain the main logic that led to a recommendation.
2. Source Attribution
   • Where possible, we cite specific standards or clauses (e.g., ISO 27001 sections, SOC 2 TSC, etc.).
   • We distinguish factual references from interpretative best practices, ensuring you know which parts of the AI’s guidance are strictly from the standard and which are suggestions.
3. Confidence Levels
   • High Confidence: When a specific standard or regulatory text explicitly states it.
   • Medium Confidence: When it’s widely accepted industry practice but not written verbatim in the standard.
   • Low Confidence: When the assistant is less certain or the situation is highly context-dependent.
4. Limitations
   • The AI may produce inaccuracies (“hallucinations”). We encourage you to verify critical advice with official documents or experts.
   • Since standards evolve, some references might become outdated over time—always check for the latest version.

3. Interpretability

Interpretability focuses on understanding how the model processes information internally. Our assistants are built on Large Language Models (LLMs), which operate as complex “black boxes.” While we cannot expose every layer of the neural network, we do:

1. Provide a Simple Reasoning Summary
   • You’ll see a brief explanation of what informed the conclusion—like referencing certain clauses or acknowledging user inputs.
   • This is not the full “chain-of-thought” but offers a high-level glimpse.
2. Acknowledge Black-Box Nature
   • LLMs are not purely rule-based. They generate responses probabilistically.
   • For highly critical or nuanced guidance, we recommend consulting a human expert or official standard documentation.
3. Encourage Validation
   • We make it easy for you to review the sources or best practices.
   • You decide whether the AI output aligns with your context, risk profile, and compliance obligations.

Feedback & Contact

• If you notice inaccurate outputs or want us to delete your conversation history, please reach out.