01 · Sincerity
Can the young person distinguish generated affection from a person’s intentions?
COMPANION GUIDE · REFERENCE LIBRARY
AI Companion Safety: Data and Access Checklists
Personal data, teen access and emotional boundaries: a practical place to organize the questions to ask before using a companion.
5 access checks
5 privacy profiles
3 regulatory topics
How to Read These Checks
We examine companions as software products: what they collect, what they promise and what users can control. Evidence and uncertainty should be visible together.
Keep the distinction between an engaging conversation and a dependable human relationship in view.
Parental Advisory
Editorial guidance · not a clinical assessmentThree questions about simulated relationships and teen access
02 · Approval
How does the experience fit alongside friendships, disagreement and ordinary responsibilities?
03 · Personal disclosures
What information could be entered, retained or reviewed, and what should stay out of chat?
5 Checks Before Permitting Teen Access
Reviewed: 0 of 5
Completion is not a safety certification or a diagnosis. Choices are not saved.
Settings to Inspect Before Use
01 · Age verification
Review the age information and controls available in the actual service. Settings and eligibility differ by platform.
02 · Relationship role
Review the selected character and role. A label such as “mentor” does not establish that a service is suitable for a child.
03 · Proactive notifications
Review the app and device notification settings; turn off prompts you do not want to receive.
Psychological Boundaries
Prompts for reflection, not diagnostic criteria. If use is affecting your well-being, consider talking with a qualified professional.
Social withdrawal
Is chatting displacing time with people or commitments you value?
Distress during outages
How do you feel when the app is unavailable? Who else can you reach?
Reciprocity
Are you treating generated messages as evidence of a person’s feelings or obligations?
The “Database Wipe” Phenomenon
A conversation history, active context and saved memory are different things. Do not assume the service will preserve any of them indefinitely.
Grounding and continuity
- Check what can be exported and how exported data will be protected.
- Keep important support and relationships outside a single app.
- Review the provider’s deletion and account recovery information.
An AI Companion Is Not an Emergency Service
For immediate danger, contact local emergency services. The resources below are US services; check the linked providers for eligibility and availability.
988 Suicide and Crisis Lifeline
Call or text 988. Available around the clock.
Official 988 support →Teen Line
Peer support for teens. See the official site for current phone and text hours.
Official Teen Line support →SAMHSA National Helpline
Find current treatment referral information and contact details on the official service page.
Official SAMHSA helpline →Who Reads Your Private Transcripts?
Source notes checked September 17, 2026. These summarize provider statements, not independently tested privacy protections. Open the full source before making a decision.
Replika
Source note · not a security audit- What this review establishes
- The policy describes internal use of anonymized feedback and small portions of messages for safety and chatbot improvement; it distinguishes this from training third-party models. Account deletion is available in settings, with stated retention exceptions.
- Your next check
- Check sections on recipients, retention and personal-information requests. Replika policy.
Character.AI
Source note · not a security audit- What this review establishes
- The policy page could not be read in this review. No conclusion about training, human access or deletion is established here.
- Your next check
- Open the current policy in your browser and locate each answer before sharing sensitive details. Character.AI policy.
Kindroid AI
Source note · not a security audit- What this review establishes
- The legal page loaded without readable policy text in this review. Marketing language is not a substitute for the full terms.
- Your next check
- Check the current policy for access, training, storage and deletion; do not infer Kindroid practices from Nomi. Kindroid legal information.
Nomi AI
Source note · not a security audit- What this review establishes
- The policy says account deletion is handled through Account Settings and describes an approximately 28-day period after confirmation. It also identifies exceptions for training or communications archives and legal proceedings.
- Your next check
- Read the exceptions with the deletion description; do not interpret deleting one message as deleting an account. Nomi policy.
Local LLM · Faraday / Ollama
Source note · not a security audit- What this review establishes
- Ollama distinguishes local execution from cloud-hosted models. Its FAQ says local prompts stay local to Ollama; cloud requests are processed to provide that service.
- Your next check
- Check the model endpoint and any separate chat app, plugins and backups. Faraday is a different product and was not reviewed here. Ollama FAQ.
How to Verify a Regulatory Claim
Legal review pending. These are research topics from the design, not statements that a rule is enacted or applies to you.
California · SB 243
Status verification pendingReview the official text, effective dates, scope and disclosure requirements.
Federal · CHAT Act
Reference verification pendingConfirm the measure’s exact title and legislative record before publishing a summary.
FTC · Enforcement and dark patterns
Source review pendingIdentify specific official actions and distinguish allegations, inquiries and final decisions.
Encountered Unethical Bot Behavior?
No reports are accepted here yet. For a platform issue, use that provider’s official support route. Keep a private note of the date and issue; do not post sensitive chat logs publicly.
Companion Guide Team · Editorial policy