Building in Public Is a Disclosure Machine

Most advice about protecting an invention warns against deliberate disclosure: don't publish the paper, don't launch the product, don't demo at the conference until you have filed. That advice assumes disclosure is a discrete event you choose. For a software team it usually is not. The work is disclosed continuously, in public repositories, changelogs, issue threads, release notes, and "building in public" posts, long before anyone stops to ask whether there was something worth protecting.

This page explains how the patent system treats that ambient disclosure, why the safety net is thinner than most builders assume, and what to decide before you publish rather than after. It is educational, not legal advice.

Last updated: July 2026. Educational only, not legal advice, and not a prediction for any specific situation. Primarily US law and USPTO practice; other jurisdictions differ and are generally stricter. Deadlines are unforgiving; consult a registered patent practitioner about your situation before, not after, you publish.

What actually counts as a disclosure

Under US law, an invention generally cannot be patented if it was "described in a printed publication" or "in public use, on sale, or otherwise available to the public" before the effective filing date (35 U.S.C. §102(a)). The phrase "printed publication" is far broader than it sounds. The test the Federal Circuit applies is public accessibility: whether the material was made available to the interested public, not whether anyone actually read it (Weber v. Provisur Technologies, 92 F.4th 1059 (Fed. Cir. 2024)).

By that standard, a great deal of ordinary development activity is a candidate disclosure:

• A public source repository that describes the technical approach, or whose code embodies it.
• An open-source release, package publication, or tagged version.
• A Show HN, Product Hunt, or launch post, and the "building in public" thread that explains the mechanism.
• A blog post, changelog, or design doc that describes how the system works, increasingly one written by an AI agent.
• A conference talk, demo, or recorded video that shows the invention in operation.
• A preprint, whitepaper, or documentation site.

No court has held that a public GitHub repository is categorically a printed publication, and accessibility is a fact question decided case by case. But a repository intended to be found, indexed, and used is exactly the kind of material the accessibility standard was written to reach. The safe assumption is that if the public can get to it and understand the invention from it, it can count.

Accessibility is only the first question, though, and clearing it does not by itself sink a patent. A disclosure matters to a particular claim only if it also teaches that claimed subject matter, and does so in enough detail that a skilled person could build it. So the analysis has three gates, not one: was the artifact accessible, what did it actually teach, and was that teaching enabling. A one-line README that names a feature clears the first gate and may fail the other two; a detailed architecture post can clear all three. Publication creates evidence. What that evidence does to a specific claim is a separate, fact-specific question for a practitioner.

One shift deserves its own mention. AI does not change patent law, but it changes how much gets disclosed and how findable it is. An agent can turn a terse commit into a detailed architecture narrative, a troubleshooting guide, or a public issue answer that explains why the system works, published automatically and indexed immediately. The disclosure risk is no longer limited to what an engineer chose to write. (Sending code to a private AI provider is a different question: it raises confidentiality and trade-secret issues, but is not by itself a public disclosure merely because an outside service received it.)

The US grace period is a safety net, not a plan

The United States is unusual in offering any grace period at all. Under §102(b)(1), a disclosure made by the inventor (or someone who got the subject matter from the inventor) within one year before filing does not count as prior art against that inventor in the US. That one year is real, and it has saved many applications.

It is also narrower and less certain than builders tend to assume, for three reasons:

• It runs against a hard deadline. The clock starts on the first qualifying public disclosure. Miss the anniversary and claims to the subject matter that artifact disclosed become barred in the US, permanently and with no cure. The deadline is tied to what was actually disclosed, not to the product in general, so later or materially different work gets its own analysis. And the date that counts is when the material became publicly accessible, not the timestamp inside a commit: code committed privately in January but pushed to a public repo in June is disclosed in June.
• Its scope is largely untested. The exception protects "the subject matter" the inventor disclosed. How closely a later reference has to match, and how much of what was published is actually shielded, are questions the courts have not thoroughly worked out. Practitioners describe the AIA grace period as far less generous than its pre-2013 predecessor. See MPEP 2153.
• It does not cleanly protect you from an independent third party. There is a related exception (§102(b)(1)(B)) under which, once you have publicly disclosed, an intervening third-party disclosure of the same subject matter may be removed as prior art. But it turns on the two disclosures matching, and it rewards disclosing first, which is precisely the behavior that forfeits your rights everywhere else. Relying on it is a race, not a strategy.

The grace period is what you fall back on when something has already gone public. It is not a reason to publish before filing.

Outside the US, do not assume you have a grace period

The US is unusual in offering a broad grace period, and a handful of countries follow suit: Canada, Australia, Japan, and Korea provide their own grace periods, typically twelve months, each with its own conditions and formalities. But two of the largest patent markets do not. Europe and China apply an absolute novelty standard: a public disclosure before the filing date destroys novelty immediately, and the ability to patent there goes with it. The European Patent Office recognizes only very narrow exceptions, such as disclosures resulting from an evident abuse or at a specific class of recognized international exhibition.

The practical consequence is stark where absolute novelty applies. A single public commit or launch post that might leave a one-year window open in the US can foreclose European and Chinese patent rights the moment it goes live. The right rule is not "you always lose every foreign right immediately," it is "do not publish before someone has identified the countries that matter and confirmed their rules," because those rules vary and change.

A separate rule points the same direction. If the invention was made in the US, a foreign-filing license (§184) may be required before filing abroad, usually granted automatically when you file with the USPTO first. That is about permission to file, not novelty, but it is another reason a US-first filing is the safe default.

"It was private" and "it was confidential" are not the same as "it was safe"

Two common assumptions deserve a direct correction.

A confidential transaction can still count. In Helsinn Healthcare v. Teva (2019), the Supreme Court held that a commercial sale to a party bound to keep the invention secret still triggered the on-sale bar under §102. The bar is not limited to public publication; a sale or even an offer for sale can start it, confidentiality notwithstanding. A confidential technical conversation, a development partnership, and a commercial offer are not the same event, though, and which of them counts is its own fact question.

Deleting the artifact does not undo the disclosure. Making a repository private, taking down a post, or force-pushing over history does not retract a disclosure that was already publicly accessible. Caches, forks, archives, and third-party copies persist. The date it first became accessible is the date that counts.

What to do instead of never shipping in public

The takeaway is not to stop building in public. It is to know which artifacts are disclosures and to decide deliberately, before they go live, rather than discovering the consequence at filing. A workable posture for a fast-moving team:

• Recognize the invention early. The protectable mechanism is usually in the non-obvious technical decision, the failure handling, the routing or verification logic, not the headline feature. If you cannot yet name what is distinctive, that is a reason to look, not a reason to assume there is nothing. See the Idea Disclosure template for a plain-language way to capture it.
• Separate the "what" from the "how" when you can. Announcing that a product exists is a different act from publishing the enabling technical detail that lets a reader build it. The second is the disclosure that matters most for novelty.
• Consider a provisional before the public artifact, not after. A provisional application establishes an early filing date and buys twelve months to a full application. Filing it before the launch post or public release is what preserves the option, including abroad. But a provisional only supports what it actually describes: a thin provisional does not preserve subject matter it never disclosed, and a nonprovisional must generally follow within twelve months to keep the benefit. See Provisional Patent Strategy.
• Track your own disclosure history. Conception dates, release dates, and public-post dates are the facts a practitioner needs to assess deadlines. A disclosure record kept as you go is far more reliable than reconstructing it from commit logs a year later. The Invention Disclosure Form leads with exactly these dates.
• When in doubt, talk to a practitioner before you publish. The one action that cannot be undone is the disclosure itself. Everything else, including deciding not to file, can wait.

Educational context on the governing framework, not legal advice, and not a prediction for any specific matter.

If something already went public

If a disclosure has already happened, the useful move is not to guess at its legal effect. It is to preserve the evidence and get it in front of a practitioner while options may still exist. Capture the URL, the repository and branch, the commit hash or release tag, the earliest date the material was publicly accessible, any later visibility changes, and the exact technical content, plus whether any application has already been filed and whether foreign markets matter. The one thing that cannot be undone is the disclosure itself. Almost everything else, including deciding not to file, can still be worked out with counsel.

Sources