A major governance crisis is unfolding in patent law: between 30 and 40 percent of practitioners now use generative AI tools for patent prosecution work, yet virtually none have updated their client engagement letters or outside counsel guidelines to address AI usage. This disconnect between adoption and disclosure represents what legal experts are calling the profession's most urgent unresolved risk.

Why Are Patent Lawyers Adopting AI So Quickly Without Disclosure Safeguards?

The pressure to adopt AI in patent prosecution stems from three converging structural forces that are reshaping the economics of the profession. Patent prosecution fees have remained flat for years, representing roughly a 50 percent real decline when adjusted for inflation. Simultaneously, the technology being patented is growing more complex, prior art databases are expanding exponentially, and the number of registered patent practitioners before the U.S. Patent and Trademark Office (USPTO) is declining. These conditions have made AI adoption not optional but structurally necessary for survival.

The efficiency gains are real and measurable. Google's A/B testing across its outside counsel panel, including major firms like Fish & Richardson, produced a 20 percent efficiency gain in patent prosecution work. In separate internal experiments, AI-generated office action response strategies matched or exceeded the strategies produced by Google's outside counsel more than 80 percent of the time.

But here's the problem: those efficiency gains are not translating into higher profits for law firms. Instead, they are creating downward pressure on fees. Google's Chief Legal Officer has directed the legal operations team to reduce outside counsel patent prosecution service fees by 30 percent, a mandate that industry observers characterize not as an isolated client demand but as a harbinger of industry-wide reorientation.

What Specific Governance Gaps Are Creating Legal Risk?

The governance gap manifests across several critical dimensions that could expose law firms to liability, ethical violations, and loss of client trust. A Berkeley-Stanford Advanced Patent Law Institute panel featuring former USPTO Director Michelle Lee, Paximal founder Ian Schick, and Google Global Patents Head of Data Science Steve Gong identified the following unresolved risks:

• Client Confidentiality Exposure: AI tools often require uploading sensitive client information to third-party vendor platforms, raising questions about privilege protection and whether vendor terms of service allow model training on proprietary client data.
• Lack of Disclosure in Engagement Letters: Virtually no law firms have updated their engagement letters to disclose AI usage, leaving clients unaware that their cases are being handled with AI assistance rather than traditional attorney analysis.
• Missing Outside Counsel Guidelines: Firms lack formal policies governing when and how AI tools can be deployed, creating inconsistent quality control and potential ethical violations across practice groups.
• Hallucination and Accuracy Issues: Google's operational data shows that AI-augmented invention disclosure forms contain hallucinations at measurable rates, yet there is no standardized process for catching and correcting these errors before they are filed with the USPTO.

The confidentiality risk is particularly acute. When patent prosecutors upload client invention disclosures, conference transcripts, or video recordings into generative AI tools to accelerate the drafting process, they may be inadvertently allowing those tools to train on proprietary client information. This creates a tension between the efficiency gains that AI promises and the fiduciary duty attorneys owe to protect client secrets.

How Should Patent Firms Implement AI Responsibly?

Industry experts have identified two competing architectural approaches to AI deployment in patent prosecution, each with different governance implications. Understanding these models is essential for firms trying to balance efficiency gains with risk management:

• Copilot Model (Human-Driven with AI Assistance): Attorneys remain in control of the drafting process and use AI as a tool to accelerate specific tasks like converting slide decks into invention disclosure forms or generating first-draft office action responses. This preserves attorney individuality but requires attorneys to develop prompt engineering skills, which are non-trivial and constantly evolving as AI models change. Quality control becomes difficult across large practice groups because different attorneys will produce inconsistent results.
• Agentic Model (AI-Driven with Human Review): AI systems encode best practices, guardrails, and defined process flows into the system itself, producing consistent output quality across the entire practice group. In this model, every user occupies the role of a senior attorney, and the AI system acts as the associate, executing multi-step drafting processes based on aligned understanding. This approach sacrifices individual attorney idiosyncrasy but ensures that junior attorneys produce work product at the same quality level as senior attorneys.
• Disclosure and Governance Framework: Regardless of which model a firm adopts, it must update engagement letters to disclose AI usage, establish clear outside counsel guidelines governing AI deployment, implement confidentiality safeguards to prevent client data from being used for vendor model training, and create quality control checkpoints to catch hallucinations and errors before they are filed with the USPTO.

"Tasking attorneys with gaining this whole new skillset that they need to be continuously trained on I think is a poor use of attorney time and attorney talent," said Ian Schick, founder of Paximal, explaining why the copilot model creates ongoing training burdens that distract from core legal work.

Ian Schick, Founder, Paximal

The agentic approach, as Schick described it, simulates the traditional partner-associate alignment process that has structured patent drafting for decades. A senior attorney reviews the disclosure and aligns on the inventive concept before the associate drafts claims and figures, with iterative alignment checkpoints. In the agentic AI model, every user occupies the senior attorney's role and the AI system takes the associate's place, executing the multi-step drafting process on the basis of the aligned understanding.

What Are the Broader Implications for the Patent Profession?

The governance gap in patent law reflects a deeper structural challenge facing the profession. New entrants to the patent bar have declined every year for approximately fifteen years, meaning the average active practitioner now has more than twenty years of experience with no comparably large cohort coming behind. Without AI-driven productivity gains, the supply of prosecution services will contract structurally even as the demand for filings grows with AI-driven innovation.

This creates a paradox: AI adoption is necessary to maintain service capacity and keep fees accessible to a broader class of inventors and innovators. But without proper governance frameworks, disclosure practices, and confidentiality safeguards, the rush to adopt AI could expose law firms to liability and damage client trust. The profession faces a critical window to establish best practices before AI usage becomes so widespread that governance gaps become industry-wide liability risks.

"The legal team is no different. The IP counsel is no different. The patent prosecutor is no different," said Michelle Lee, former USPTO Director, noting that boards of directors are asking their C-suite leaders to demonstrate measurable return on investment from AI implementation across all functions, including patent prosecution.

Michelle Lee, Former USPTO Director, Obsidian Strategies

For patent practitioners, the message is clear: AI adoption is inevitable, but governance is not. Firms that proactively update their engagement letters, establish clear AI usage guidelines, implement confidentiality safeguards, and create quality control checkpoints will be better positioned to capture efficiency gains while managing legal and ethical risks. Those that continue to deploy AI without disclosure or governance frameworks are gambling with client trust and professional liability.