The following explanatory notes regarding the limitations of AI in patent searching and patent drafting were generated using ChatGPT (OpenAI), based on general industry practice and publicly available regulatory principles.

Why AI Cannot Be Used to Write or File Patent Applications

AI presents significant risks when used for drafting, preparing, or filing patent applications. These risks affect confidentiality, legal validity, claim quality, and patentability.

Below are the key issues.

1. Confidentiality & Disclosure Risks

a. Entering an invention into a public AI system may be treated as disclosure

When a client types their invention into a public AI tool:

the data passes through third-party servers,
the provider may store or log the information,
automated internal systems may access it.

This breaks confidentiality and can be interpreted as disclosing the invention to a non-bound third party, potentially jeopardizing novelty.

b. Inventions must remain confidential until filing

Patent systems (especially the U.S. “first-to-file” rule) require the invention to remain confidential until a patent application is filed.
AI systems cannot guarantee that.

c. AI providers are not bound by NDAs, CPATA rules, or professional standards

A patent agent/matter is confidential by law.
AI is not.

2. Legal Drafting Risks

a. AI cannot perform legal analysis

Drafting claims, determining essential features, and defining scope require legal judgement.
AI cannot:

determine patentable subject matter,
evaluate statutory requirements,
assess novelty/non-obviousness,
construct defensible claim language.

b. AI often creates legally invalid or unenforceable claim sets

Common failures include:

missing essential elements,
inconsistent terminology,
claims that do not support the description,
improper dependencies,
indefiniteness,
ineligible subject matter.

These errors can destroy an otherwise valid invention.

c. AI “hallucinates” facts

AI may invent:

prior art references,
technical features,
definitions that do not exist.

In patent drafting, fabricated details can:

introduce unsupported material,
create internal contradictions,
result in rejection or later invalidation.

3. Patent Office Compliance Issues

a. Patent offices require human accountability

USPTO, CIPO, EPO all require:

a legally responsible human applicant or agent,
verifiable assertions,
accountability for misstatements.

AI cannot:

certify accuracy,
assume liability,
attest to inventorship or declaration statements,
provide sworn documentation.

b. AI is not recognized as an inventor or drafter

Patent law requires human inventorship.
If AI appears to have contributed materially to drafting or conception, this can create inventorship issues or even invalidate a filing.

Agencies have issued guidance stating that AI cannot be named as an inventor or contributor.

4. Technical Drafting Problems

a. AI cannot ensure enablement and written description

A valid patent requires:

complete description,
workable embodiments,
feasible processes,
support for every claimed feature.

AI often produces:

vague embodiments,
missing steps,
unworkable designs,
overly broad or inadequately supported descriptions.

b. AI is poor at identifying the “inventive concept”

A human must:

isolate what is new,
articulate why it is non-obvious,
distinguish from prior art.

AI lacks the reasoning structure needed to do this reliably.

5. Ethical and Professional Responsibility Issues

a. CPATA, USPTO, and EPO require competence

Patent agents and attorneys must:

exercise professional judgement,
perform due diligence,
maintain confidentiality.

AI cannot meet competence standards for legal practice.

b. Clients receiving AI-drafted patents face real legal risks

A poorly written application can cause:

permanent loss of patent rights,
weak claim protection,
inability to enforce the patent,
expensive prosecution corrections,
rejection or invalidation years later.

6. Strategic Issues

a. AI drafts often reveal the invention prematurely

Even if the draft is never filed, simply generating it via a public AI may:

expose key elements,
compromise novelty,
undermine international filing rights.

b. AI misses strategic claim opportunities

A good patent is not just a technical document.
It is a legal strategy document crafted to:

block competitors,
maximize scope,
anticipate design-arounds,
future-proof the invention.

AI cannot think that way.

Why AI Cannot Be Used for a Formal Patent Search

A formal patent search must meet professional, legal, and confidentiality standards that current AI systems cannot fulfill. Here are the key reasons:

1. Confidentiality Risks (Most Important)

a. Public-facing AI tools are not private

When an invention concept is entered into most commercial AI systems, the information may be:

processed on external servers,
temporarily stored for system performance,
accessed by internal automated systems for model training or improvement (depending on the platform’s policies).

Even when a platform claims not to train on user inputs, the data still passes through third-party infrastructure.

This creates a risk of accidental disclosure, which is unacceptable before a patent is filed.

b. Loss of novelty = loss of patent rights

If sensitive invention details are disclosed to a third party (including an AI system on external servers), this can count as:

public disclosure, or
loss of confidentiality,

which can directly jeopardize patentability under “first to file” systems.

A human searcher working under an internal NDA does not create this risk.
AI systems do.

2. AI Cannot Access or Search Official Patent Databases Properly

A formal search requires deep review of:

USPTO PAIR / Patent Center
CIPO databases
WIPO PATENTSCOPE
EPO Espacenet
International classifications (CPC/IPC)
Non-patent literature databases

AI systems do not have direct access to these databases. They rely on:

incomplete training data,
inconsistent coverage,
snapshots of old patent information,
no guarantee of up-to-date filings.

Patent data updates daily; AI models cannot track that in real time.

A formal search requires live, database-accurate, classification-based searching. AI cannot perform this.

3. AI Cannot Interpret Claims or Legal Scope Reliably

Patent searching is not keyword searching.
A formal search requires human judgement to interpret:

claim scope,
inventive concept,
embodiment coverage,
analogous prior art,
combining multiple references.

AI currently:

misinterprets claim construction,
misses contextual equivalents,
cannot reliably distinguish essential vs. optional features,
“hallucinates” prior art that does not exist.

Therefore, an AI-generated search result is not legally dependable.

4. AI Outputs Are Not Auditable or Defensible

A formal patent search must be:

documented,
reproducible,
defensible,
verifiable,

for later use during prosecution or in litigation.

AI systems cannot provide:

search paths,
classification trees,
logic chains,
database queries,
review notes.

Because AI output is generated probabilistically and not based on traceable queries, it is not admissible or defensiblein a patent context.

5. AI May Expose the Invention to Unknown Third Parties

Submitting an invention to a public AI service exposes it to:

unknown subcontractors,
cloud providers,
model trainers,
automated quality systems.

These entities are not:

bound by a confidentiality agreement,
regulated by CPATA,
subject to professional liability.

A patent search requires a controlled confidentiality environment.
AI systems cannot guarantee that.

6. AI Cannot Replace Structured Human Search Methodology

A formal patent search requires:

CPC/IPC classification analysis,
backwards and forwards citations,
inventor and assignee filters,
non-patent literature review,
manual relevance ranking,
search term expansion using controlled vocabularies.

AI has no ability to perform these structured steps in a professional, verifiable manner.

AI can sometimes assist indirectly (e.g., brainstorming keywords), but it cannot perform the actual formal search.