A Compliant Path To AI For Patent Attorneys And Corporate IP Teams

I have been working in the field of artificial intelligence patent drafting for a long time. My experience in this area began well before the development of Large Language Models. From the beginning, I have observed strong resistance to the idea of automated patent drafting. This resistance has continued since the first day I started promoting the concept.

The past three years have been particularly intensive and demanding for me. During this time, I have gained several important insights, which I would like to share below.

Based on my long-standing experience, I have developed a plug-in for Microsoft 365, and more specifically for Copilot Studio. We call that plug-in "Invention Harvesting Copilot" (click here). The article that follows provides a detailed explanation of why I decided to develop this tool.

The Compliance Wall That Blocks AI

Artificial Intelligence is already transforming the way patents are searched, drafted, and managed. Yet many patent professionals cannot use AI at all in their daily work.

The reason is not a lack of willingness or skill. It is the legal and organizational framework in which they operate.

For patent attorneys, professional secrecy rules prohibit the use of any system that might expose client data to third parties. For corporate patent departments, IT governance and data-protection policies block access to most public AI tools.

Both groups share the same question: How can we use AI securely, without being in a danger of getting accused of breaking the laws and ethical duties that define our profession?

From Early Experiments to a Secure Solution

I began working on the automation of patent drafting long before today's large language models existed. My first experiments were done in Excel with Visual Basic and later through a collaboration with the Technical University of Darmstadt. We attempted to automate certain elements of claim drafting by analysing structural features from gearbox design.

Years later, during the early stages of the COVID-19 period, I began to explore how large language models could serve the same purpose. This led to collaborations with several AI technology providers.

The real turning point came with the Symposium on Robot Patent Drafting, which I co-organised in Cannes on 22–23 September 2022. For the first time, patent attorneys, technologists, and industry experts discussed both the benefits and the dangers of automated patent drafting. The discussions were published as a book (click here) that remains relevant today, because it captures the earliest hopes and concerns surrounding AI in patent work.

Those debates made one thing clear: AI can only be used responsibly in our field if it operates in an environment that guarantees security, confidentiality, and professional compliance.

Training 200 Patent Professionals – What the Experience Revealed

After the symposium, I started to train patent attorneys and patent engineers in the use of AI-based drafting tools. Over time, I conducted around 200 individual training programmes with professionals from law firms and industry.

The results were measurable. On average, users doubled their drafting speed. The best performers achieved four to eight times their previous productivity. About one fifth of the participants struggled, often not because of the technology itself, but because of concerns about data handling and compliance.

This experience taught me an important lesson. The main barrier to AI adoption in patent work is not technical. It is institutional trust. AI systems must meet the same level of reliability, traceability, and legal safety that the profession demands of every other tool we use.

The Four Fears That Stop AI Adoption in Patent Work

When I speak with patent attorneys and in-house patent engineers, the same four concerns appear again and again:

1. Data Theft – Fear that external AI providers might access or leak confidential invention data, especially in defence or pharmaceutical sectors.

2. Cross-Contamination – Concern that client or company data used to train an AI system could resurface in another user's output.

3. Public Disclosure – Risk that using a cloud-based AI platform could be considered a "public disclosure," creating prior art against a later patent filing.

4. Breach of Secrecy – The gravest concern for patent attorneys, who must protect client information under criminally enforceable professional secrecy laws.

These fears are not theoretical. They reflect real ethical and legal boundaries that cannot be ignored in patent practice.

Professional Secrecy and Confidentiality — The Foundation of Trust

Patent work depends on confidentiality. Every client relationship and every invention is protected by law or by internal policy. Yet the exact legal framework differs for law firms and for corporate patent departments.

For patent attorneys in private practice, the relevant rule is the duty of professional secrecy, also called the professional secrecy obligation. It is defined in the Patentanwaltsordnung and in the professional conduct code of the European Patent Office. This duty forbids any disclosure of client information to third parties. Breaches can lead to disciplinary or even criminal consequences.

For patent professionals in companies, the equivalent standard is found in corporate confidentiality requirements. These are anchored in employment contracts, nondisclosure agreements, and IT-security frameworks. They ensure that trade secrets, research data, and invention disclosures remain under strict company control.

Both legal frameworks aim at the same goal: to protect sensitive information and preserve trust.

Our Invention Harvesting Copilot was created precisely to fulfil these obligations. It allows users to apply AI-based methods without crossing the boundary of professional secrecy or corporate confidentiality. All data remain inside the company's or firm's Microsoft 365 tenant, and no information is transferred to external servers.

The result is a system that increases efficiency without compromising legality, ethics, or trust.

The Common Dilemma in Law Firms and Companies

Despite clear benefits, both groups face the same technical obstacle. Many Patent attorneys do not want to use Software-as-a-Service AI tools because doing so could breach their professional secrecy obligation. Corporate patent engineers often are not allowed to use them because their IT departments forbid unverified cloud connections.

Installing an internal large language model (LLM) might appear to be a solution, but it is costly, complex, and rarely maintained effectively. The result is that both private practitioners and corporate teams remain excluded from the productivity gains that AI could deliver.

The Key Insight – Trust the System You Already Use

The breakthrough came from a simple observation. Almost every law firm and every corporate patent department already uses Microsoft 365 for its daily confidential work. Its ecosystem has already been approved by IT departments and by professional associations. It is certified under ISO 27001, SOC 2, and GDPR standards, and Microsoft offers clear legal recourse through local entities in almost all countries of this world.

This meant that the infrastructure for compliant AI use already existed. What was missing was a specialised AI interface that worked inside the safe Microsoft tenant, not outside it.

The Solution – Our Invention Harvesting Copilot

To meet this need, we developed the Invention Harvesting Copilot, a plug-in for Microsoft Copilot Studio. The tool integrates seamlessly into Microsoft Teams and SharePoint and functions entirelywithin the user's existing Microsoft 365 tenant. No data ever leaves the internal environment.

We keep things simple by using the Invention Harvesting Copilot for collecting

How the Invention Harvesting Copilot works:

1. The inventor and the patent professional meet in Microsoft Teams, another one of the many apps of the Microsoft 365 ecosystem.

2. The conversation is transcribed automatically, and later uploaded into the Invention Harvesting Copilot app.

3. Drawings and a reference signs list are uploaded into the Invention Harvesting Copilot app.

4. The Invention Harvesting Copilot app produces a structured draft invention disclosure that can be reviewed and finalised immediately.

The system is fully compliant with both professional secrecy obligations and corporate confidentiality requirements. Because it operates inside the existing Microsoft 365, no separate IT approval or special security exception is required.

Deployment and Integration

Deployment is simple.

The plug-in is installed once inside the tenant with cooperation from the user's IT department. The process takes only a few hours.

Licensing is per tenant, not per user. Whether a company has one user or several thousand, the annual subscription remains the same.

No external servers, tokens, or API accounts are needed. The tool functions with Microsoft's own Copilot Studio environment, which ensures legal and data-protection compliance by design.

Practical Benefits for Both Groups