Core Principle

Large Language Models (LLMs) are stateless reasoning engines. They do not serve as systems of record and do not retain memory beyond a single interaction.

Information Never Provided to LLM Context

• Passwords, API keys, tokens, session cookies, or cryptographic secrets
• Full government identifiers or full financial account numbers
• Complete medical records or raw diagnostic datasets
• Unfiltered inboxes, full call recordings/transcripts, or entire documents
• Instructions authorizing payments, access, or irreversible actions
• Internal security enforcement logic not intended for disclosure

What May Be Provided

• Minimum necessary task context
• Normalized summaries and redacted excerpts
• Non-sensitive metadata and references (IDs/pointers)

Persistence

Persistent data is stored in secured systems outside the LLM (databases, encrypted storage, auditable logs). The LLM does not store memory independently.

Decision Authority

AI outputs are advisory. Final decisions and actions are governed by system rules, validation layers, and human approval where required.