China’s ACPAA Bans AI-Generated Patent Filings, Limits Fee Competition

China just sent the patent world a message that is about as subtle as a gavel on a glass table: yes, artificial intelligence matters, yes, AI innovation is booming, and no, that does not mean patent agents get to hand the steering wheel to a bot and call it a day. The All-China Patent Attorneys Association, or ACPAA, has revised its professional ethics rules to ban patent agents from using AI to directly generate patent application documents for final submission. It also tightened the screws on unfair pricing by targeting the malicious undercutting of patent agency fees.

At first glance, those changes may look like two separate housekeeping items. In reality, they tell one bigger story about where China’s intellectual property system is headed. The country still wants more high-value innovation, especially in AI-related technologies. But it also wants better patent quality, more reliable professional accountability, less junk in the filing pipeline, and fewer bargain-basement service wars that turn serious patent work into a race to the cheapest keyboard.

For companies, inventors, and in-house legal teams, that means China is not turning against AI. It is turning against careless AI use in patent drafting and against business models that treat patent prosecution like discount fast food. The menu is still open. The mystery meat is what got banned.

What ACPAA Actually Changed

The headline change is straightforward: patent agents may not use artificial intelligence to directly generate patent application documents that are ultimately submitted. That rule is important because it goes straight to the heart of professional responsibility. A patent filing is not supposed to be a glorified autocomplete exercise. It is a legal-technical document that must accurately describe an invention, preserve claim scope, support later enforcement, and survive scrutiny from examiners, courts, and competitors who would love nothing more than to poke holes in it.

The second change focuses on fee competition. ACPAA’s revised code does not outlaw ordinary price competition or negotiated fees. What it targets is malicious price cutting in patent agency services as part of broader unfair competition rules. In other words, the issue is not that agencies must all charge the same amount. The issue is that firms should not slash fees so aggressively that they distort the market, mislead clients, hollow out professional standards, or create pressure to churn out low-quality filings at industrial speed.

That pairing matters. When a regulator or industry body cracks down on AI-generated patent documents and predatory fee competition at the same time, it is usually saying one thing in two dialects: quality matters, and quality costs human effort.

Why China Is Taking a Harder Line

1. Confidentiality risks are not theoretical anymore

Patent drafting begins with technical disclosure. That often includes unpublished product plans, prototypes, manufacturing methods, algorithm details, chemical formulas, testing results, and internal engineering notes. Feed that material into the wrong AI tool and the legal problem is not just embarrassment. It can become a trade secret leak, a novelty problem, a security incident, or a malpractice headache with extra paperwork on top.

Chinese authorities have been increasingly explicit about that risk. In 2026, CNIPA warned against the use of AI agents such as OpenClaw for drafting patent application documents, citing dangers like excessive permissions, security vulnerabilities, malicious plugins, and leakage of core technical information. That warning did not appear out of thin air. It followed earlier concern in 2025 from Nanjing’s Intellectual Property Protection Center, which barred AI-generated content in patent application documents submitted for pre-examination.

Put plainly: if the invention walks into the AI tool and the confidentiality walks out the back door, nobody gets a gold star.

2. Patent drafting is still a judgment job, not a paste job

Anyone who has spent five minutes with a generative AI system knows the magic trick. It sounds confident, writes fast, and occasionally invents things like a novelist on espresso. That is charming when you are brainstorming blog headlines. It is less charming when you are drafting claims for a semiconductor architecture, a biotech platform, or a machine-learning pipeline that needs precise support in the specification.

Chinese regulators have flagged that problem too. The concern is not just obvious hallucinations. It is also subtle technical defects: contradictions in logic, vague feature descriptions, overbroad language, unsupported embodiments, and mismatches between the inventive concept and the final claims. Those flaws can weaken patent protection or sink an application entirely.

That is why the ACPAA rule is best read as a professional accountability rule. The patent agent who signs the filing remains responsible for the filing. “The AI did it” is not likely to become a winning defense in any serious jurisdiction, and it certainly is not the kind of sentence you want floating around a quality audit.

3. China wants fewer bad-faith filings and more high-value patents

China’s patent system has been evolving from a quantity-heavy model toward a quality-focused one. That policy direction shows up across official messaging about high-value patents, examination quality, and the need to discourage bad-faith or low-substance applications. AI-generated patent documents create an obvious temptation in that environment: faster drafting, lower cost, more filings, and maybe a few extra statistical trophies. Regulators seem to have decided that this is exactly how filing systems get clogged with weak applications that waste everyone’s time.

And here is the nuance that makes the story more interesting: China is not anti-AI in patents. Not even close. CNIPA has issued guidance for AI-related inventions, updated examination standards, and continued signaling that real AI innovations can be patentable when they involve genuine technical solutions and comply with law, ethics, and public-interest requirements. China also remains a global heavyweight in AI patent activity. So the policy is not “AI bad.” The policy is much closer to “AI innovation welcome, AI ghostwriting not welcome.”

Why the Fee Competition Rule Matters Almost as Much

On paper, “maliciously lowering the price of patent agency services” may sound like the less flashy change. In practice, it may be just as important. Patent drafting is labor-intensive when it is done well. It requires inventor interviews, claim strategy, prior-art awareness, translation accuracy where needed, disclosure calibration, and a thousand tiny judgment calls that determine whether a filing becomes a real business asset or an expensive wall decoration.

When agencies compete mainly on rock-bottom fees, corners get cut. Maybe the inventor interview gets shorter. Maybe the claims get copied from an old template. Maybe junior staff are stretched too thin. Maybe AI tools start doing more than assisting and quietly begin substituting for professional analysis. Before long, the client thinks they bought efficient prosecution when what they really bought was a stylishly formatted risk.

The ACPAA’s fee competition restriction therefore works hand in hand with the AI-drafting ban. One targets the production method. The other targets the business pressure that can push firms toward that method. Together, they nudge the market away from volume-first behavior and toward a more defensible value proposition: thoughtful drafting, better quality control, and less temptation to flood the system with bargain-bin applications.

China’s Position Compared With the United States

This is where the comparison gets especially useful for multinational companies. In the United States, the USPTO has taken a more permissive approach to AI-assisted patent practice. It has made clear that AI-assisted inventions are not categorically unpatentable, so long as a natural person made a significant contribution to the invention. The agency has also said the use of AI-based tools in practice is not prohibited. But there is a catch, and it is an important one: practitioners remain responsible for accuracy, candor, confidentiality, and compliance with existing rules.

That U.S. position has inspired a practical middle ground. AI can help with brainstorming, summaries, drafting support, prior-art review, formatting, and workflow efficiency. But humans must still own the inventive contribution, verify the output, protect confidential information, and make the legal judgments that actually matter.

China’s new ACPAA rules do not erase that middle ground entirely, but they draw the line sooner and more sharply when it comes to final filing documents. The contrast is revealing. The United States is saying, “Use AI carefully.” China, at least in this slice of patent practice, is saying, “Use it if you must, but do not let it write the final paper you are signing.”

Neither system is embracing fully autonomous patent drafting as the new normal. The difference is one of tolerance and timing. China appears less willing to let the market run a giant live experiment on filing quality while everyone pretends the chatbot has excellent bedside manner.

What This Means for Businesses Filing in China

For foreign applicants

If you file patents in China through local counsel, expect more questions about drafting workflow, human review, and document provenance. Companies that rely heavily on generative AI in early drafting stages may need tighter review steps before China filings move forward. This is especially true where multilingual drafting, translation, and claim adaptation are involved.

For Chinese patent agencies

Agencies will likely need clearer internal controls, stronger confidentiality policies, and training on what counts as prohibited direct generation versus acceptable support use. They may also need to rethink pricing models that depend on speed and scale at the expense of document quality.

For in-house legal teams

This is a good moment to audit AI policies. Ask which tools are being used, what data enters them, whether prompts are retained, whether providers train on user inputs, who reviews outputs, and whether the final drafting decisions are being made by qualified humans. If that sounds tedious, welcome to patents. Precision has always had a paperwork side hustle.

For inventors and startups

The cheap-and-fast promise of AI-drafted patent applications may look tempting, especially when budgets are tight. But a weak patent can cost more than no patent at all if it creates false confidence, collapses under examination, or proves impossible to enforce. The new ACPAA rules are basically a reminder that patent filings are not just documents you submit. They are legal foundations you may someday have to stand on during a storm.

The Bigger Picture: China Wants Better Patents, Not Just More Patents

The real significance of the ACPAA update is that it captures a broader shift in modern IP policy. Patent systems everywhere are wrestling with the same tension: AI can increase speed and reduce cost, but it can also amplify error, dilute accountability, and tempt users into outsourcing judgment. China’s answer, at least for patent filing practice, is to protect the system from that slide before it becomes routine.

That should not be mistaken for technological conservatism. China is still deeply invested in AI, still encouraging AI-related inventions, and still occupying a leading global position in AI patent output. But its regulators and professional bodies are showing that pro-innovation does not mean pro-shortcut. They want human responsibility attached to the legal act of filing, and they want pricing norms that do not reward reckless throughput.

For the global patent bar, the lesson is simple enough to fit on a sticky note: AI may speed up the drafting process, but it cannot replace the professional duty behind the signature. And when a patent office ecosystem starts worrying about both robot-written filings and cut-rate fee wars, it is usually because it has seen what happens when quality gets treated like an optional add-on.

That tends to end badly, usually with more rejected claims, more preventable disputes, and at least one very awkward meeting where someone says, “But the tool sounded so sure of itself.”

What Teams Are Experiencing on the Ground

Across companies that file in China, the lived experience around these rules is becoming more practical and less theoretical. Legal teams are discovering that AI policy is no longer something to discuss at innovation conferences with shiny slides and free coffee. It is now a workflow question. Who drafts what? Which tools are allowed? What gets uploaded? Who reviews the final claims? And perhaps most importantly, who will take responsibility when a filing turns out to contain language that looks polished but does not actually protect the invention?

One common experience is the shock of seeing how persuasive weak AI output can look on first read. A startup founder or engineer may open a draft and think, “This seems great.” Then outside counsel reads it and spots the problem parade: missing fallback positions, unsupported embodiments, inconsistent terminology, vague technical effects, or claims that sound broad but collapse the moment prior art enters the room. That experience tends to convert AI enthusiasm into AI caution very quickly.

Another recurring experience involves confidentiality. In-house teams often realize, a little later than they would like, that not every AI product handles sensitive data the same way. Once that happens, internal discussions get serious. Teams start asking whether invention disclosures were entered into public or lightly secured tools, whether prompts were retained, and whether internal approval rules were followed. Suddenly, “convenient drafting assistant” starts sounding suspiciously like “possible information leak with legal accessories.”

Patent agencies are experiencing a different pressure. Many clients still want speed and lower costs, and some assume AI should automatically deliver both. At the same time, regulators and professional bodies are making it clear that final responsibility stays with the practitioner. That puts agencies in the middle of a delicate balancing act. They want efficiency, but not at the price of ethics violations, quality failures, or reputational damage. As a result, many firms are rethinking internal review ladders, approval checkpoints, training requirements, and the exact point where AI assistance stops and human professional judgment must take over completely.

Foreign businesses filing in China are also learning that rules can differ meaningfully across jurisdictions even when the technology is the same. A workflow that might be acceptable in one country with strong human review may create problems in another if local regulators are less tolerant of AI-generated filing text. That experience is pushing multinational IP teams toward more jurisdiction-specific governance instead of one-size-fits-all global patent playbooks.

And then there is the pricing experience. Clients who once chased the lowest filing quote are starting to appreciate why ultra-cheap patent work can carry hidden costs. A bargain application that lacks precision may invite office actions, narrowing amendments, enforcement weakness, or expensive refiling strategy later. In that sense, the ACPAA’s warning about malicious fee competition reflects something many sophisticated filers learn eventually: in patents, the discount often arrives first, and the invoice for the real damage shows up later.

Conclusion

China’s ACPAA has done more than tweak an ethics code. It has drawn a clearer boundary around professional patent practice in the AI era. Patent agents cannot simply let artificial intelligence generate final filing documents and press submit like they are ordering lunch. Agencies also cannot lean on destructive fee competition and pretend quality will somehow survive the squeeze. Together, those rules point to a more disciplined patent market: one that still welcomes real AI innovation, but expects humans to stay accountable for the legal work that turns inventions into enforceable rights.

For businesses, the takeaway is refreshingly unglamorous and therefore extremely useful. Use AI with care. Protect confidential disclosures. Keep humans in charge of the final drafting judgment. Document inventorship clearly. And treat suspiciously cheap patent services the way you would treat suspiciously cheap parachutes: with respectful distance and a few follow-up questions.