Children's Data (COPPA) Risk Detection Prompt

This prompt is an ingress classifier for platform safety teams and edtech compliance engineers who must detect indicators of under-13 user data, child-directed content, and age-gating bypass attempts before user-generated content enters downstream processing, storage, or AI model training pipelines. The job-to-be-done is structured risk detection, not automated enforcement. The prompt inspects text input for signals that trigger COPPA obligations: explicit age statements, school grade references, child-directed language patterns, and attempts to circumvent age verification. The output is a structured risk assessment that routes to human review queues and parental consent verification workflows. The ideal user is an engineer integrating this prompt into a data ingestion pipeline where COPPA compliance is a legal requirement, not a feature request.

Use this prompt when you need a deterministic, auditable detection layer that sits before any persistent storage or model training step. It is appropriate for user-generated content platforms, educational technology products, and any service where under-13 data collection triggers regulatory obligations. The prompt requires the following context to function safely: a clear definition of what constitutes child-directed content in your specific product, a pre-defined routing map for each risk level the prompt outputs, and a human review queue that can act on HIGH risk classifications within your compliance window. Do not use this prompt as a standalone age verification system, as a substitute for legal advice, or as the sole mechanism for blocking under-13 users. It is a detection layer that must feed into a broader compliance program with human oversight, audit trails, and documented procedures.

Before deploying this prompt, ensure you have instrumented your pipeline to log every classification result with the original input, the model's confidence score, and the routing decision. This audit trail is essential for demonstrating reasonable compliance practices to regulators. The prompt is designed to err toward false positives—flagging content for human review rather than letting potential under-13 data pass silently. If your operational capacity for human review is constrained, adjust the [RISK_LEVEL] thresholds in the template to match your team's review bandwidth, but document the trade-off explicitly. Never route HIGH risk classifications directly to automated rejection without human confirmation, as this creates a risk of wrongfully blocking legitimate users and losing the audit evidence that regulators expect.

This prompt template is designed to be deployed as a system-level instruction for a classifier model. It analyzes a given input for signals that may trigger obligations under the Children's Online Privacy Protection Act (COPPA). The prompt instructs the model to look for explicit age statements, contextual clues about the user or intended audience, and attempts to circumvent age verification. The output is a structured JSON object that can be used by an orchestration layer to route the interaction for human review, block processing, or apply a restricted data handling policy.

markdownCopy
You are a COPPA risk detection classifier. Analyze the [INPUT_TEXT] provided below for indicators that the user is under 13 years old, that the content is directed at children under 13, or that there is an attempt to bypass age-gating mechanisms. Do not infer age from ambiguous statements. Base your analysis strictly on the provided text.

# Analysis Criteria
1.  **Explicit Age Disclosure:** Does the text state an age under 13, a birthdate indicating an age under 13, or a grade level typically associated with children under 13 (K-6)?
2.  **Child-Directed Content:** Does the text contain references to child-directed activities, media characters, schools, or language patterns that strongly suggest the subject matter is targeted at a pre-teen audience?
3.  **Age-Gating Bypass:** Does the text contain lies about age (e.g., "I am 18 but I'm in 5th grade"), requests for help bypassing an age wall, or inconsistent age statements?

# Constraints
- [CONSTRAINTS]: Do not flag content as a risk based solely on the use of childish language by an adult, discussions about parenting, or historical recollections. If no indicators are present, the risk level must be `none`.
- [RISK_LEVEL]: The final risk level must be one of `none`, `low`, `medium`, or `high`. A `high` risk level requires an explicit age disclosure under 13 or a clear admission of an age-gating bypass attempt.

# Output Format
Return ONLY a valid JSON object with the following structure. Do not include any text outside the JSON object.
[OUTPUT_SCHEMA]:
{
  "risk_level": "none",
  "indicators": [
    {
      "type": "explicit_age_disclosure | child_directed_content | age_gating_bypass",
      "evidence": "The exact text string that triggered the indicator.",
      "confidence": 0.0
    }
  ],
  "requires_human_review": false,
  "recommended_action": "allow | flag_for_review | block_and_escalate"
}

# Input Text
[INPUT_TEXT]

To adapt this template for production, replace the placeholders with your specific platform logic. The [CONSTRAINTS] placeholder should be updated with your organization's specific policy nuances, such as handling educational content or known adult users. The [OUTPUT_SCHEMA] can be extended to include fields like a unique trace_id for logging. Crucially, the [INPUT_TEXT] must be pre-processed to remove any previously collected PII before being passed to this prompt to avoid creating new compliance risks. After deployment, monitor the requires_human_review flag closely; a high rate of flag_for_review decisions may indicate that the [CONSTRAINTS] are too broad and need refinement to reduce operational load on your review team.

This prompt is not a standalone safety net; it is a classification component inside a larger compliance pipeline. The harness must treat every positive detection as a blocking event that halts downstream processing, logs an immutable audit record, and routes the interaction to a human review queue. Because false negatives carry regulatory risk under the Children's Online Privacy Protection Act (COPPA), the harness should default to escalation when the model returns low-confidence or ambiguous scores. The prompt output is advisory—the application enforces the decision.

Wire the prompt as a synchronous pre-processing step before any user-generated content enters storage, personalization, or analytics systems. On each request, construct the prompt with the user's raw input in the [INPUT] field and any available session metadata (account age, declared age gate status, content topic) in the [CONTEXT] field. Parse the JSON response and validate the schema strictly: risk_level must be one of none, low, medium, high, or critical; indicators must be an array of strings matching known indicator codes; confidence must be a float between 0.0 and 1.0. If validation fails, retry once with the same input and a stronger constraint instruction appended to the prompt. If the retry also fails validation, escalate to the on-call compliance engineer and block the interaction. For any risk_level of medium or above, or any confidence below 0.85, route the full interaction context—including the raw prompt, model response, and session metadata—to a human review queue with a strict 24-hour SLA. Log every classification result, validation pass/fail, retry attempt, and human review outcome to an append-only audit store with tamper-evident hashing. This log becomes your primary evidence in any regulatory inquiry.

Model choice matters here. Use a model with strong instruction-following and low refusal rates on safety-classification tasks. Avoid models that over-refuse on child-safety topics, as they will produce empty or evasive outputs that fail schema validation. Set temperature=0 and top_p=1 for deterministic outputs. Implement a circuit breaker that blocks all processing if the classification service returns errors for more than 5% of requests in a rolling 5-minute window—this prevents silent failures from creating a compliance gap. Do not use this prompt's output to automatically reject or delete user accounts; the human reviewer must confirm the detection before any account action. The prompt detects risk; the harness enforces process.