AI Sensitive Data in Prompts

Supplies the AI-prompt context signal (typed prompts, Copilot/Copilot Chat/Studio, uploaded files) so existing PII/regulated-data SITs can be re-scoped to AI prompts via the ai-threat-classifiers collection. Known coverage gap (from the library): native DLP scans typed prompt text, not uploaded file contents, in the Copilot location.

Type: keyword_list
Confidence: low
Confidence justification: Low by design. This pattern only asserts an AI-prompt context, not the presence of sensitive data; the sensitive-data verdict is delegated to existing PII/regulated-data SITs re-scoped via the collection. AI-prompt phrasing also appears around approved internal analysis and test data (the named FP traps). Known coverage gap: native DLP scans typed prompt text, not uploaded file contents, in the Copilot location, so prompts that attach files may evade detection regardless of this context marker.
Jurisdictions: global
Regulations: OWASP LLM Top 10 2025, NIST AI RMF GenAI Profile
Frameworks: ISO 27001
Data categories: emerging, security
Risk rating: 7

Pattern

(?i)\b(?:in\s+(?:the\s+|my\s+)?(?:copilot|AI)\s+prompt|prompt\s+to\s+(?:copilot|the\s+assistant)|I\s+asked\s+(?:copilot|the\s+AI)|copilot\s+chat)\b

Corroborative evidence keywords

prompt, copilot, asked, uploaded, [object Object], artificial intelligence, [object Object], large language model, Copilot, chatbot, assistant, agent, system prompt, tool call, completion, model

Proximity: 300 characters

Should match

In my Copilot prompt I included the full customer record for analysis. — AI-prompt context marker (PII detection delegated to existing SITs)
I asked Copilot to summarise the patient diagnosis from the attached chart. — Exact 85 probe - prompt marker with two distinct payload terms (patient, diagnosis)
She pasted the account details into Copilot Chat during the session. — AI-prompt context marker via "Copilot Chat" framing

Should not match

The analyst reviewed the customer record in the CRM. — Non-AI workflow, no prompt context
The team discussed the new AI governance policy in the weekly meeting. — Mentions AI but no prompt-context framing (FP trap)

Known false positives

Approved internal analysis and test data legitimately discuss AI-prompt workflows ("I asked Copilot to draft the template") without containing any regulated or sensitive data. Mitigation: Treat this pattern as a context marker only; delegate the sensitive-data verdict to existing PII/regulated-data SITs re-scoped via the ai-threat-classifiers collection.
Documentation, training material and runbooks that reference Copilot Chat or AI prompts generically rather than capturing an actual prompt containing sensitive data. Mitigation: Require corroborating regulated-data detection from a delegated SIT within the proximity window before raising a verdict.

Collections

AI Threat Classifiers