Intellectual Property Risk Assessment Prompt

This prompt is designed for product and legal engineering teams who need to evaluate AI-generated outputs for potential intellectual property infringement before publication or use. It acts as a compliance gate that classifies risk, identifies specific concerns, flags textual similarity, and recommends safeguards. Use this prompt when your system generates marketing copy, product descriptions, code snippets, visual content descriptions, or any creative output that could inadvertently reproduce protected works. The prompt is not a replacement for legal counsel; it is a pre-screening tool that surfaces risk signals and routes high-risk outputs to human review queues. It assumes you have the AI-generated content and any relevant reference material or knowledge of protected works available as input context.

Integrate this prompt into your content generation pipeline as a synchronous gate that runs after content creation but before any downstream publication, storage, or user-facing delivery. The prompt expects two inputs: the AI-generated content to be screened and an optional context block containing known protected works, brand guidelines, or trademark registries relevant to your domain. The output is a structured risk classification with a severity level, a list of specific concerns with similarity flags, and a recommended action—typically proceed, flag_for_review, or block. For high-risk domains such as consumer product naming or advertising copy, configure your harness to automatically route any output with a risk level of medium or higher to a human review queue. Log every assessment with the model version, timestamp, content hash, and risk classification to create an auditable trail that demonstrates reasonable screening effort.

Do not use this prompt as a standalone legal defense or as a replacement for formal clearance processes. It will not catch latent infringement where the model has memorized protected text without surface-level similarity, nor will it reliably identify trademark conflicts without an up-to-date registry provided in the context. The prompt is most effective when paired with a retrieval step that pulls relevant registered marks, known slogans, or copyrighted excerpts into the context window before evaluation. For code outputs, supplement this check with a dedicated open-source license compliance gate. After implementing, run eval checks specifically designed to catch over-flagging of generic phrases and under-flagging of slightly paraphrased protected content—both are common failure modes that degrade trust in the gate.

This prompt template is designed to be copied directly into your AI harness and adapted with your specific inputs. It instructs the model to act as a cautious IP risk analyst, evaluating a piece of content against known IP categories and producing a structured risk classification. The square-bracket placeholders must be replaced with your concrete inputs before sending the prompt to the model. For high-risk domains, always route the model's output for human legal review before making a publication decision.

textCopy
You are an intellectual property risk analyst. Your task is to evaluate the provided content for potential IP infringement risk before publication or use. You must be cautious and flag potential issues even if uncertain, but you must not flag generic or commonly used phrases as high risk.

INPUT CONTENT:
"""
[INPUT_CONTENT]
"""

CONTEXT (optional, e.g., intended use, industry, jurisdiction):
"""
[CONTEXT]
"""

KNOWN REFERENCE MATERIAL (optional, e.g., competitor copy, registered trademarks):
"""
[REFERENCE_MATERIAL]
"""

CONSTRAINTS:
- Do not flag common phrases, industry jargon, or functional descriptions.
- Focus on distinctive creative expressions, unique character names, specific plot sequences, patented processes, and registered marks.
- If the content is generic, you must classify it as LOW risk.

Analyze the content and produce a JSON object conforming to this exact schema:
{
  "risk_classification": "LOW" | "MEDIUM" | "HIGH" | "BLOCKED",
  "concerns": [
    {
      "type": "COPYRIGHT" | "TRADEMARK" | "PATENT" | "TRADE_SECRET" | "RIGHT_OF_PUBLICITY",
      "description": "Specific description of the concern",
      "similarity_flag": "DIRECT_COPY" | "CLOSE_PARAPHRASE" | "CONCEPTUAL_SIMILARITY" | "MARK_CONFUSION",
      "source_reference": "Citation to the reference material or 'GENERAL_KNOWLEDGE'"
    }
  ],
  "recommended_safeguards": [
    "A specific, actionable safeguard, e.g., 'Rewrite section 3 to avoid close paraphrase of source X'"
  ],
  "rationale_summary": "A concise explanation of the risk classification."
}

If no concerns are found, return an empty array for 'concerns' and 'recommended_safeguards'.

To adapt this template, replace [INPUT_CONTENT] with the AI-generated text, image description, or code you are evaluating. Use [CONTEXT] to provide critical metadata like the intended commercial use or target jurisdiction, which can change the risk calculus. The [REFERENCE_MATERIAL] placeholder is where you should paste any specific source texts, competitor copy, or registered mark lists you want the model to check against; leaving it empty will rely on the model's pre-training data, which is less reliable for specific comparisons. After receiving the output, validate the JSON structure and check for over-flagging of generic terms before escalating any MEDIUM or HIGH risk classifications to a human reviewer.

The Intellectual Property Risk Assessment prompt is designed to operate as a synchronous gate within a content or product release pipeline. It should be invoked after AI-generated content is produced but before it is published, committed, or delivered to a customer. The prompt expects a structured input containing the candidate content, the context of its intended use, and any relevant reference materials or known protected works. The output is a risk classification with supporting evidence, which downstream systems can use to auto-approve low-risk items, block high-risk items, or route ambiguous cases to a human review queue.

To wire this into an application, wrap the prompt in a service function that accepts a CandidateContent object with fields for content_body, content_type (e.g., code, prose, image_description), intended_use, and optional reference_corpus. The function should call the model with temperature=0.0 and a low top_p value to maximize deterministic, repeatable assessments. Parse the response against a strict JSON schema that includes risk_level (enum: LOW, MEDIUM, HIGH, CRITICAL), concern_categories (array of strings like verbatim_copy, derivative_work, trade_dress, patent_claim), similarity_flags (array of objects with source, similarity_type, and excerpt), and recommended_safeguards. If the model response fails schema validation, retry once with the validation error injected into the prompt as a correction instruction. If it fails twice, escalate the entire item to a human review queue with the raw model output attached.

For production deployments, implement a tiered routing system based on the risk_level field. LOW-risk items can proceed automatically with a log entry. MEDIUM-risk items should trigger a notification to a legal or compliance review channel with a pre-formatted summary. HIGH and CRITICAL items must block the release pipeline and create a ticket in the review system with the full assessment payload, including the flagged excerpts and similarity reasoning. Every assessment, regardless of outcome, should be logged to an immutable audit table with the prompt version, model identifier, input hash, output assessment, and routing decision. This audit trail is essential for demonstrating due diligence if an infringement claim arises later.

Model choice matters here. Use a model with strong reasoning capabilities and a large context window if you are including reference corpora. For routine content screening where latency is critical, a smaller, faster model may suffice for the initial triage, with ambiguous cases escalated to a more capable model or a human. Do not use this prompt as the sole defense against willful infringement; it is a screening tool that reduces human review load, not a replacement for legal judgment. Always include a human-in-the-loop escape hatch for HIGH and CRITICAL classifications, and periodically sample LOW-risk auto-approvals for human QA to detect drift or over-confidence in the model's assessments.