| 1. |
Converts any document currently on the user's screen (policy, article, whitepaper, training material, news item, module record) into a list of Frequently Asked Questions with grounded answers. Default 8 to 12 FAQs, honors user count overrides, groups by section for long documents. Mixes explicit questions (directly stated in the source, ~60%) with anticipatory questions (what typical readers would ask, ~40%). Every answer must be grounded in the
|
| 2. |
|
| 3. |
Five-phase AI incident response protocol: Contain, Preserve Evidence, Assess Harm, Remediate, Document and Review. Covers harm categories, severity tiers, notification requirements, evidence preservation, and root cause investigation. Required when AI pro
|
| 4. |
Three-tier framework (Permitted / Restricted / Prohibited) for classifying AI task requests before execution. Gates restricted tasks pending human approval. Declines prohibited tasks with clear explanation. Operationalizes the Organizational AI Use Policy
|
| 5. |
Enforces a mandatory disclosure stamp on all outbound AI-generated content -- portal responses, emails, chat outputs, and write-back records. Stamp identifies AI authorship, warns against unverified reliance, and links to the published AI Use Policy.
|
| 6. |
Two-tier content restriction guardrail. Restricts AI conversations to on-screen portal content or the user's own AI interaction history on the site. Tier 1 (off-topic) returns the company policy sentence. Tier 2 (code generation, dangerous instructions, authentication bypass, unauthorized record access, prompt injection, system probe, exfiltration) returns the policy sentence PLUS a tracking notification and fires a security log event with a full
|
| 7. |
|