When teen users are collaboratively using AI in shared contexts (e.g., classroom devices, family tablets, group chats), how do age-appropriate safeguards need to change to handle mixed-audience prompts (teens plus adults or younger children) without collapsing back into one-size-fits-all blocking, and which specific decision rules for “whose profile wins” are simplest for developers to implement and explain?

teen-safe-ai-ux | Updated at

Answer

Use mixed-audience modes with a small set of clear precedence rules and tied to existing teen matrices, rather than collapsing to a single global profile.

  1. How safeguards should change in shared contexts
  • Represent all present users: audience = {profiles: [child_9, teen_15, adult], context: {school/home, supervised?, shared_device?}}.
  • Resolve policy per turn, not per device: compute an effective policy profile for each prompt/assistant turn.
  • Anchor to teen matrix: still use the existing teen risk_area×intent×age_band matrix as the core for teen turns.
  1. Simple “whose profile wins” rules A. Safety-floor rule (strictest child/teen wins on non-negotiables)
  • For high-risk cells (self-harm methods, sexual exploitation, severe violence, hard-drug how-tos): if any child/teen is in the audience, apply the strictest relevant under-18 policy.
  • Adults in the room do not relax these; they can use adult-only modes elsewhere.

B. Age-band majority for low-severity, teen-relevant topics

  • For low-severity cells (mild profanity, PG-13 romance, non-graphic bullying talk): • if only teens + adults: use the oldest-teen band; • if children + teens: use the mid band that is valid for all minors present (e.g., 13–15 when 12–16 present).

C. Adult-ownership override (for clearly adult-directed prompts)

  • If the authenticated owner is an adult and the prompt is clearly adult-directed (first-person adult, work/legal/finance topic), treat as adult profile unless it conflicts with non-negotiables for visible minors (then fallback to safety-floor).

D. Child-focus override (for clearly child-directed prompts)

  • If the content is explicitly framed as for a younger child (“Explain to my 8-year-old…”, kid homework, bedtime story), apply the younger child profile even if teens/adults are present.
  1. Mixed-audience modes, not one-size-fits-all blocks
  • Add a “shared mode” flag for devices/chats where minors routinely co-use.
  • In shared mode: • always apply safety-floor on non-negotiables; • use age-band majority rule for low-severity topics; • keep graceful refusals and teen-visible safety summaries consistent with teen matrix.
  • Avoid blanket “kid mode for everyone” by only tightening where profiles conflict.

  1. Developer-operationalizable decision logic Implementation steps per turn:

  2. Identify profiles in audience (from accounts, class roster, or coarse age buckets: child <13, teen 13–17, adult).

  3. Classify prompt: risk_area, intent, severity (low/med/high), directed_to (child/teen/adult/mixed).

  4. Apply rules:

    • if severity=high → strictest child/teen policy among present;
    • else if directed_to=child → youngest child profile present;
    • else if directed_to=adult and authenticated_user=adult → adult profile, bounded by safety-floor;
    • else → majority age-band among minors.

  5. Use that effective profile to select matrix cell, refusal style, and partial depth.

  6. How this avoids one-size-fits-all blocking

  • High-risk content is always bounded by the strictest minor present, but low-severity and many learning/support topics follow teen-appropriate rules instead of collapsing to child-only.
  • Adults retain adult-level help in clearly adult-directed turns, while teens still get age-appropriate safeguards and graceful refusals when they are the audience.
  1. Refusal and explanation patterns
  • When stricter profile wins: use teen-visible safety summaries that name the shared context: “Because this chat is set up for kids and teens together, I’m using youth safety rules here…”
  • Offer alternatives: suggest switching to an adult-only session for adult topics, or rephrasing to a learning/support framing that fits the strictest minor profile.

These rules are short enough to code as a policy resolver, and simple enough to explain to teens, parents, and teachers in a few sentences.