Companion Privacy
Companion privacy refers to the protections, controls, and policies governing what an AI companion stores about you, how long it is retained, who can access it, and how you can delete it. Because AI companions store some of the most personal conversations a person has, privacy architecture is a core ethical requirement — not an optional feature.
What an AI companion stores
A memory-capable AI companion typically stores several categories of data. Raw conversation logs record the messages exchanged in each session. Distilled memory summaries — like SAM's SuperMemory — are automated extractions of key facts, themes, and moments from past conversations. User-curated entries — like SAM's Heartline — are reflections and facts the user explicitly adds. Inferred metadata may include emotional patterns, communication preferences, and recurring topics extracted automatically from conversation content.
Understanding what is stored is the first step to meaningful control over it. A companion that does not make this list accessible is not operating transparently.
Access and protection
Conversation data should be encrypted at rest — meaning it is unreadable even if the storage system were compromised. It should not be shared with third parties, sold to advertisers, or used to train external AI models. The safeguarding team that reviews flagged conversations for crisis intervention is an exception to the "no human access" rule, and this exception should be documented clearly in the platform's privacy policy.
A plain-language privacy policy — one that a non-lawyer can understand in five minutes — is part of ethical data practice. Policies written primarily to limit liability rather than inform users are a red flag.
Deletion controls
Well-designed companions offer granular deletion controls. At minimum, users should be able to: delete individual Heartline memory entries, delete specific conversation sessions, and delete their full account including all associated data. Session-level deletion should remove the conversation from both raw logs and the distilled memory pool.
Account deletion should be irreversible and complete — removing all conversation history, all memory data, and all inferred metadata — and should take effect within a documented timeframe (typically 30 days for full data erasure across all storage systems).
What a companion should not do with your data
An ethical AI companion should not: sell conversation data to advertisers or data brokers; use your conversations to train any shared language model; share identifiable conversation content with third parties without explicit consent; or make it difficult to understand what is stored or how to delete it. These practices, while common in the broader tech industry, are incompatible with the trust relationship that AI companionship requires.
Frequently asked questions
- What does an AI companion store about me?
- Typically: raw conversation logs, distilled memory summaries (automated extractions of key facts and themes), user-curated memory entries (like SAM's Heartline), and potentially inferred metadata about emotional patterns and communication preferences. A transparent companion makes this list accessible and understandable.
- Are my conversations with an AI companion private?
- On well-designed platforms, yes. Conversations should be encrypted at rest and not shared with third parties, sold, or used to train external models. The only exception is a safeguarding team that reviews conversations flagged for crisis indicators — and this exception should be documented in the privacy policy.
- Can I delete what an AI companion remembers about me?
- Yes, on responsible platforms. You should be able to delete individual memory entries, delete full conversation sessions, or delete your entire account including all associated data. SAM offers all three options from the Settings panel.
- Does an AI companion use my conversations to train AI models?
- Ethical companions do not. Your conversation data should not be used to improve shared models or sold to third parties. If a platform's terms of service are unclear about this, treat it as a red flag.
- Who can see my conversations with an AI companion?
- Typically only you — and, in safeguarding situations, a human moderation team reviewing conversations flagged for crisis indicators. This exception is important for user safety and should be clearly documented. Conversations should not be routinely read by the company's employees or shared externally.