Is Data Leakage the Biggest AI Risk for Patent Drafting? Think Again.

When we talk about Generative AI, the conversation almost always starts with: “What about data leakage?” But relying on fear might be creating a blind spot.

Here is my take on AI risk in patent practice:

  1. Model Training: Early AI models trained on user inputs, but mainstream providers now offer environments where “training on user data” is disabled by default. The days of accidentally training ChatGPT with your trade secrets are mostly behind us.

  2. Client Confidentiality: It’s About Transparency. If your firm uses cloud storage, you are already entrusting data to third parties. Most enterprises are results-oriented. As long as the quality is high and security is vetted, the “how” matters less than the transparency.

💡 For Firms: Consider adding disclaimer clauses regarding AI tools to manage expectations. 💡 For Enterprises: Require firms to disclose AI usage and ensure their vendors adhere to strict confidentiality standards.

  1. Hallucinations: The REAL Danger. This is where I lose sleep. Unlike data leakage, which is a security issue, hallucinations are a competence issue. A made-up experimental result isn’t just embarrassing. It’s a breach of fiduciary duty and potentially fatal to the patent.

My takeaway: AI is here to stay. We need to shift our energy from “blocking” to “verifying.” The future belongs to those who can effectively fact-check the AI output.