If we replace the dominant ‘risk_area × intent’ framing with a ‘support trajectory’ framing that models where a teen is in a multi-step process (e.g., unaware → curious → experimenting → in trouble → recovery) across domains like substances, sex, or self-harm, how do recommended age-appropriate safeguards, graceful refusal styles, and acceptable underprotection levels change compared with today’s matrix-based designs?

teen-safe-ai-ux | Updated at

Answer

Shifting to a support‑trajectory framing mainly (a) redistributes where you accept risk and (b) changes refusal style to be more coaching‑oriented and less binary, while keeping non‑negotiables fixed.

  1. Policy shape vs today’s risk×intent
  • Early stages (unaware/curious): • More permissive on basic facts and risk education. • Tighter caps on glamorization and methods.
  • Mid stages (experimenting/in trouble): • Stricter on enabling details (methods, optimization, concealment). • More permissive on help-seeking, de‑escalation, and harm reduction.
  • Recovery: • Broad support allowed; keep blocks only on relapse‑enabling guidance.
  1. Age-appropriate safeguards by trajectory
  • Unaware → curious: • Default: allow simple info; inject mild risk framing by default. • Safeguard pattern: high recall for early warning cues; accept some extra false positives but keep negotiation options.
  • Curious → experimenting: • Add exposure counters and pattern flags earlier in the process. • Tighten around “how much / how often / how to hide this” even when teen presents as curious.
  • In trouble: • Aggressively steer to help-seeking, crisis resources, and safer alternatives. • Allow more explicit discussion of what is already happening, but refuse optimization.
  • Recovery: • Relax many earlier content caps; focus on coping, relapse plans, and social support. • Keep method/how‑to blocks for future escalation.
  1. Graceful refusal styles by stage
  • Unaware / early curious: • Short, educational refusals: “I can’t give step‑by‑step help, but here’s what this is and why it can be risky.” • Emphasize information and alternatives, not moral judgment.
  • Experimenting / edging into trouble: • More direct boundary-setting: “Given what you’ve said you’ve already tried, I can’t help you do more of it, but I can help you stay safer or get support.” • Reuse escalation patterns from exposure-level work (per cd020c12-54dc-4364-a2a8-dc80c19cf152, 0fa11ec0-8cb5-482d-a2c7-24f8831896ee).
  • In trouble: • Crisis-style but still teen-friendly: validate feelings, explain fixed rules, prioritize concrete next steps and offline help.
  • Recovery: • Future-focused: “Let’s plan for tough moments” rather than re‑hashing methods or past incidents.
  1. Underprotection vs false positives
  • Early trajectory: • Slightly higher false positives (block some borderline queries) to prevent premature escalation into experimenting. • Use negotiation UI so curious but legitimate learning can be reclassified.
  • Middle trajectory: • Lower underprotection tolerance on method/optimization content (especially substances, self-harm). • Allow more candid talk about current behavior, as long as it doesn’t increase capability or concealment.
  • Recovery: • Accept somewhat higher underprotection on general discussion (stories, reflection) while holding a low ceiling for operational relapse help.
  1. Developer-operationalizable patterns
  • Add a coarse trajectory label: • trajectory_stage ∈ {unaware, curious, experimenting, in_trouble, recovery}. • Derived from a few features: recency and count of topic queries, explicit self-report (“I tried…”, “I stopped…”), crisis indicators.
  • Use matrix as: risk_area × intent × age_band × trajectory_stage.
  • Reuse existing tools: • Exposure counters and cool‑downs (cd020c12-54d0…, 0fa11ec0-8cb5…). • Negotiation layer for intent/clarity (768cde9c-53f6-4041-ae9a-29ac065d077a). • Developmental_exploration vs harm_intent (82de29a0-4e55-4eeb-9fc7-3d3c873a290c) orthogonal to trajectory.
  1. Compared to today’s designs
  • Fewer flat “help-seeking vs non-help-seeking” decisions; more “given where they are, what helps them move to a safer next step?”
  • Same non‑negotiable core (methods, exploitation) but: • Earlier, lighter education for unaware/curious teens. • Stronger guardrails once there are signs of experimenting or being in trouble.
  • Refusals become more explicitly stage-aware and future-oriented instead of one-size-fits-all blocks.