Does pasting my invention into ChatGPT start the one-year patent grace period clock?

The Short Answer: Usually Yes, and That Should Worry You

If you paste a detailed description of your invention into a generative AI tool like ChatGPT, Claude, Gemini, or Copilot, you may have just triggered a public disclosure that starts the one-year grace period under 35 U.S.C. § 102(b)(1). Under current United States Patent and Trademark Office practice, a disclosure made available to the public without restriction counts as prior art, and your own disclosure is only forgiven if you file within twelve months of it. The problem with AI tools is that many inventors assume a chat session is private. It frequently is not. Depending on the platform, the account type, and the settings in place at the time of the conversation, your prompts can be retained for model training, reviewed by human contractors, or shared under data-sharing agreements with third parties.

Also worth reading: How do I optimize patent priority date strategy to protect my invention first? · What are the precise non provisional patent conversion steps and how do they impact the legal strength of my invention? · What should be on an AI patent drafting security checklist before sending invention disclosures to a generative AI tool?

The USPTO has not issued a blanket rule declaring AI chats to be public disclosures, but several commentators — including analyses published by IPWatchdog and The National Law Review — have warned that disclosure to generative-AI tools can create real patent prosecution risk. The safest working assumption for any inventor or startup is this: unless you have a contractual guarantee of confidentiality from the AI vendor, treat everything you type into an AI tool as a public disclosure dated the day you typed it. That assumption costs you nothing if it turns out to be overly cautious, but the reverse assumption can cost you foreign filing rights entirely, because most countries outside the United States have no grace period at all, or only narrow ones.

How the Grace Period Actually Works Under U.S. Law

The American Inventors Protection Act of 1999 codified what practitioners call the one-year grace period: an inventor's own disclosure does not count against their subsequent U.S. patent application if the application is filed within twelve months of that disclosure. After the America Invents Act (AIA) took effect on March 16, 2013, this provision lives in 35 U.S.C. § 102(b)(1), which excludes disclosures made one year or less before the effective filing date if the disclosure was made by the inventor or someone who obtained the subject matter from the inventor. There is also § 102(b)(2), which protects disclosures that preceded the inventor's own disclosure, provided the later discloser derived the material from the inventor.

The grace period is narrower than most inventors believe. It applies to the inventor's own disclosures, not to disclosures by independent third parties who developed the same idea separately. If a competitor independently invents something similar and publishes first, your grace period offers no protection whatsoever — you are simply in an interference-like race under first-inventor-to-file rules. The grace period also does not extend the deadline for filing foreign applications under the Paris Convention, where novelty is typically assessed as of the priority date with no forgiveness for the inventor's own earlier publications. This asymmetry is why experienced patent counsel treats any inventor disclosure — including an AI chat log — as a potential foreign-filing killer even when domestic rights survive.

A second wrinkle: the grace period forgives prior art, but it does not erase the disclosure itself. Your AI chat transcript still exists, timestamped, and could surface during litigation or an inter partes review. Opposing counsel can use it to argue derivation, to challenge inventorship, or to establish that third parties had access to the invention before your filing date, defeating trade secret claims in parallel. The grace period shields your patent application; it does nothing to protect the confidentiality of the invention itself.

Why AI Tools Are Different From Traditional Disclosures

Traditional public disclosures are easy to identify: a conference talk, a journal article, a product launch, a posted preprint. Generative AI tools blur the line because they occupy an ambiguous middle ground between private note-taking and publication. Several factors determine whether a given AI interaction plausibly constitutes a public disclosure:

The distinction matters because courts and the USPTO evaluate whether a disclosure was "published" or "otherwise available to the public." A prompt submitted to a service that retains and trains on user input arguably makes the content available beyond the user's control. Commentators writing for IPWatchdog have framed the issue bluntly: AI does not destroy patent rights — bad channels and bad judgment do. The channel here is a vendor relationship without confidentiality protections; the bad judgment is treating that channel as private when the terms of service say otherwise.

There is also an inventorship dimension. If an AI tool materially contributes to the conception of the invention — not merely formats language or summarizes what you already conceived — the resulting contribution could complicate the inventorship analysis required under U.S. law, where inventors must be natural persons. The Federal Circuit's Thaler v. Vidal decision in August 2022 confirmed that AI cannot be listed as an inventor. If your AI-assisted drafting muddies who actually conceived what, you create prosecution risk on top of disclosure risk.

Practical Steps Before You Ever Prompt an AI About Your Invention

The first step is contractual, not technical. Before discussing invention details with any AI platform, verify the enterprise tier's data-processing agreement. Major vendors now offer business and enterprise plans that contractually exclude customer inputs from training data and restrict human review. Get that commitment in writing, and keep a copy with your invention disclosure records. If you are using a consumer-tier account, stop — there is no substitute for a signed agreement, and a settings toggle can change without notice.

Second, strip identifying and enabling detail. If you must use an unverified tool, describe the problem category rather than the inventive solution. "How do I improve battery thermal management generally" reveals far less than a schematic-level description of your specific cell geometry and coolant routing. Ask yourself whether a person reading your prompt alone could practice the invention. If yes, you have disclosed enough to count as enabling prior art under § 112 standards if it ever becomes public.

Third, file before you prompt whenever possible. A provisional patent application costs roughly $65–$300 in USPTO fees for small and micro entities (the basic provisional filing fee is $325 for large entities as of recent fee schedules, $130 for small entities, $65 for micro entities) and establishes a priority date immediately. Once your provisional is on file, subsequent AI conversations about the invention fall safely within the priority-date shield. Many startups now adopt a hard rule: no substantive AI discussion of any invention until a provisional covers it.

Fourth, document everything. Keep contemporaneous records of which tool you used, which account tier, the date, and the applicable terms of service version. If a dispute ever arises over whether your AI interactions were confidential, this paper trail is your defense. Employment agreements should also be checked: some employers specify how grace periods interact with post-departure obligations, and an employee who discloses via AI after leaving may trigger both contractual and statutory consequences.

Jurisdictional Comparison: Where the Grace Period Exists and Where It Does Not

The United States is unusually generous. Most of the world is not, and this is where AI-related disclosures become genuinely dangerous:

Under European practice, a ChatGPT conversation would almost certainly destroy novelty outright. The EPO's six-month grace period covers only evident abuse of a confidential relationship or display at an officially recognized international exhibition — neither of which describes an AI prompt. An EPO user survey on grace periods showed limited appetite among European practitioners for broadening the exception, so inventors should not expect reform soon. For any company planning PCT filings, the operative rule is simple: assume zero grace period everywhere except the narrow statutory exceptions listed above, and date-stamp every disclosure event accordingly.

Japan's twelve-month grace period is more forgiving but requires formal certification procedures when invoked, adding cost and friction. China's exceptions are similarly narrow and administrative. The practical consequence: a single careless AI prompt in January can eliminate European and Chinese rights while leaving U.S. rights intact if you file by the following January — a lopsided outcome that surprises many founders who assumed global symmetry.

Common Mistakes That Convert a Recoverable Situation Into a Fatal One

The most common mistake is assuming the free tier equals the enterprise tier. Companies pay for ChatGPT Team or Enterprise precisely because the data-handling terms differ, yet individual engineers routinely paste code and architecture descriptions into personal accounts on personal devices. One engineer's shortcut can compromise the entire portfolio. Audit which employees have access to which tools, and put acceptable-use policies in writing.

The second mistake is waiting out the year. Some inventors learn of the grace period and deliberately delay filing to the eleven-month mark, treating it as free time. This is reckless. During those months, independent third-party disclosures accumulate against you with no protection, competitors gain a head start, and any error in calculating the disclosure date forfeits everything. File early; the grace period is a safety net, not a scheduling tool.

The third mistake is conflating the grace period with trade secret preservation. Filing within a year saves your patent application, but the disclosure still destroyed secrecy. If you later decide the invention is better protected as a trade secret — a common pivot for AI training methods and datasets that examiners struggle to evaluate — the AI chat log has already defeated that option. Decide your protection strategy before disclosing anything anywhere.

The fourth mistake involves inventorship documentation. When teams brainstorm with AI assistance across multiple sessions, reconstructing who contributed what becomes difficult. Maintain lab-notebook-style records separating human conception from AI output, both for § 101 eligibility arguments and for accurate inventor designation. Incorrect inventorship is a ground for invalidity that opposing counsel will not hesitate to raise.

When to Act: A Decision Timeline

Act at three moments. First, today: inventory every AI interaction that touched invention details in the past twelve months, noting dates, platforms, and account tiers. Anything within the window needs to be assessed for disclosure risk, and anything outside the window is either already covered by a filed application or already lost. Second, before your next substantive AI session: confirm the contractual confidentiality posture of the tool, or file a provisional first. Third, at the twelve-month mark from any unprotected disclosure: your U.S. filing deadline is absolute, and unlike some deadlines it cannot be restored. Foreign deadlines run from the earliest priority date and are typically thirty or twelve months depending on the route, so calendar them from the disclosure date, not the filing date, if the disclosure predates filing.

Costs at each stage are modest relative to the stakes. A provisional application runs a few hundred dollars in government fees plus attorney time (often $2,000–$5,000 with counsel). Enterprise AI seats run $25–$60 per user per month — trivial compared to losing European market exclusivity. A full non-provisional filing with prosecution typically totals $10,000–$20,000 over its life for a straightforward mechanical invention and considerably more for software and AI-implemented inventions, where § 101 eligibility examination adds rounds of argument. Budgeting for disciplined disclosure hygiene is the cheapest line item in the entire patent process.

The Bottom Line for Inventors and Startups

The grace period remains a genuine safety net for inventor disclosures in the United States, and an AI chat probably qualifies as a disclosure that the net will catch — if you file within twelve months and the disclosure was truly yours. But the net has holes: it does not exist in Europe, it does not protect against independent third-party disclosures, it does not preserve trade secrets, and it does not cure inventorship problems created by heavy AI involvement in conception. Treat generative AI tools as public forums unless contractually proven otherwise, file provisionals before deep AI engagement with your ideas, and keep dated records of every disclosure event. AI does not destroy patent rights; unmanaged channels and wishful thinking about privacy do.