Operating · Lesson 20 — What public disclosure does to your patent rights
O20Operating
Operating · Lesson 20● live

What public disclosure does to your patent rights

Pitches, write-ups, and consumer-LLM prompts can each be disclosures. Recognize them before they happen.

12 min read · 60 min applycompanion: Security & IP Discipline

What disclosure means in patent law (operator’s read)

I’m an operator, not a lawyer. I’ve had to make the call on this enough times that I learned the rules the way you learn rules when getting them wrong is expensive. Here’s the operator’s read.

In patent law, a public disclosure is any non-confidential conveying of the invention to someone outside the company. Publishing a paper is a disclosure. So is a trade-show demo. So is posting on a public repo. So is pitching investors without a signed NDA. And — this is the one most founders miss — pasting proprietary method detail into a consumer-tier LLM is also a disclosure, because the terms on those tools allow training and don’t guarantee confidentiality.

The bar is lower than the instinct. The instinct says “disclosure” means press release. The doctrine says disclosure is anythingnon-confidential. Showing your method to a friend who didn’t sign an NDA is a disclosure. So is a conversation at a conference. So is a single screenshot in a public Slack.

Why it matters: the moment a disclosure happens, your patent clocks change. In the US you get a one-year grace period to file. In most of the rest of the world — Europe, China, most absolute-novelty jurisdictions — you lose the right to patent in those countries the moment the disclosure goes public. That second one is the one that quietly kills most early-stage IP strategies.

The story

The first time I really internalized this, I was about to give a demo to a venue partner. I’d been talking to them under a soft handshake — no NDA in place because we’d known each other a while and it felt like overkill. The demo was going to be in their facility, in front of their staff, including a couple of vendors they worked with.

I caught it about a week before. Vendors I didn’t know, no confidentiality framework, novel method, no filing on file. Walking into that room would have been a public disclosure to people I couldn’t bind retroactively. US grace period starts immediately; foreign rights gone the moment any of those vendors stepped out the door.

The fix was small. Signed NDAs in place 48 hours before the demo with every named attendee. A provisional patent application filed three days before. Total cost: a few hours of paperwork and ~$130 in USPTO fees. The cost of having skipped that step is harder to quantify because foreign patent rights, once gone, are gone — there’s no salvage path.

After that, the rule got mechanical. Nothing about a patentable method goes public — including to anyone outside the cap table — until a provisional is on file and NDAs are signed.That’s a checkpoint, not a vibe.

Three quiet disclosures

The three disclosure paths that don’t feel like disclosure. Hover any card to see what to actually do.

01

The investor pitch without an NDA

claim looks likeFounder takes a meeting, walks through the method on a few slides, leaves with "great chat, let us think about it."
what’s missingWithout a signed NDA in place before the meeting, that pitch is a public disclosure. The US one-year grace period starts ticking. Europe and China — absolute-novelty jurisdictions — are already gone if anything in those slides was novel and patentable.
the moveNDA before any meeting that touches the method. If the investor refuses to sign (some won't), strip the deck to outcomes and market — nothing that reads on a claim. Save the method walk for after a signed NDA or after a provisional is on file.
02

The consumer-LLM prompt

claim looks likeEngineer pastes the novel pipeline into ChatGPT to debug a step. "Just helping me think through it."
what’s missingConsumer-tier LLMs ingest the prompt under terms that allow training and don't guarantee confidentiality. Courts have already started treating this as voluntary disclosure — in *Trinidad v. OpenAI* a trade-secret claim was dismissed because the secret was developed inside ChatGPT. *United States v. Heppner* held that sharing with a commercial AI tool waived attorney-client privilege.
the moveProprietary method detail only ever touches enterprise-tier tools with no-training data terms, signed DPA, and zero-retention. Everything else uses scrubbed inputs — describe the shape of the problem, not the secret sauce.
03

The public write-up

claim looks likeFounder posts a beautifully clear technical blog post explaining how the system works. Lands on Hacker News. Great for hiring.
what’s missingAlso great for nuking foreign patent rights. The post is a published disclosure on a fixed date. US: one-year clock starts. EPO, China, most of the world: gone the moment the post went up if you hadn't filed first.
the moveProvisional first, write-up second. The cost of filing a provisional before a public post is ~$130 in USPTO fees plus a few hours of drafting. The cost of skipping that step is your entire international patent footprint.

Common thread: none of these feel like “publishing.”The pitch is private. The LLM prompt is private. The blog post is just marketing. But each one is non-confidential transfer to outsiders — that’s the doctrinal definition of disclosure, regardless of how it feels.

The grace-period trap

The US has a one-year grace period: from the date of your first public disclosure, you have 12 months to file a US patent application. That’s generous, and it’s the reason a lot of early-stage US founders never feel the pain — they keep their disclosures domestic and file within the year.

The trap shows up the moment you want to file internationally. Europe and China run on absolute novelty. If you publish, demo, or otherwise disclose before filing, you lose the right to patent in those jurisdictions immediately and permanently. There’s no grace period to recover into. Canada and Japan have limited grace periods, but they typically require filing within one year of the disclosure rather than within one year of your priority date — which is a meaningfully tighter constraint than the US version.

The practical implication: if international patent rights matter to your business — which is almost always the case for hardware, consumer products, B2B SaaS sold to global customers, anything with a global market — then the order matters. File first, disclose second. Reversing those two steps forfeits the foreign filings.

The cheapest version of “file first” is a US provisional patent application. It costs ~$130 in USPTO fees for a micro-entity. It buys you a priority date that’s recognized by most foreign patent offices under the Paris Convention. From the moment the provisional is on file, you can disclose freely — pitches, demos, papers, posts — and still have 12 months to convert to a non-provisional and to file foreign equivalents off your priority date.

If a method is going to be patentable and global, the math almost always favors the provisional. The cost of filing is much smaller than the cost of losing foreign rights. The decision rule is: before any non-confidential disclosure that touches a novel method, a provisional is on file.

Annotated transcript

A real conversation — operator about to demo at a venue, hasn’t filed anything yet, walking through what counts as disclosure and what to do before the demo. Click any pinned annotation for commentary.

Pre-demo disclosure check

Four operator turns, four agent turns. Result: clear definition of disclosure, the consumer-LLM rule, and the order-of-operations fix.

  1. operator

    I've got a pilot venue who wants to demo the system to their customers next month. I haven't filed anything yet. Am I about to do something dumb?

  2. agent

    Probably yes — depends on what the demo shows.

    A public demo where outsiders can see the system operate is a public disclosure. In the US that starts a one-year clock — you have 12 months from the demo date to file a non-provisional. In Europe, China, and most of the rest of the world, that demo permanently kills your patent rights in those jurisdictions because they run on absolute novelty.

    The right move before the demo is to file a provisional application. ~$130 in USPTO fees, gives you a priority date, locks in 12 months of "patent pending" while you decide whether to convert. Once the provisional is on file, the demo is safe — the disclosure happens after your priority date.

  3. operator

    What actually counts as "public" — does it need to be on stage with a camera?

  4. agent

    No. Public, in patent terms, means non-confidential. A few things that all count as disclosure:

    1. Publishing a paper, blog post, or README that explains the method.
    2. Demoing at a trade show, conference, customer site, or anywhere outsiders can observe.
    3. Posting code in a public repo where the method is recoverable from the source.
    4. Pitching investors without a signed NDA — even one-on-one, even in a coffee shop.
    5. Sending the method to a consumer-tier LLM as a prompt. The terms on most consumer tools allow training, which means the data is no longer confidential to you.

    The rule is: if anyone who hasn't signed a confidentiality agreement could plausibly know the invention from what you said or showed, it's a disclosure.

  5. operator

    What about my own engineers? If I tell them, is that a disclosure?

  6. agent

    No — as long as they're under an employment agreement with a confidentiality clause (every standard one has this). Employees are bound to confidentiality by default. Same goes for contractors, but only if there's a signed CDA/NDA — never assume.

    The failure mode here is contractors who started before paperwork was finalized, or "advisors" you talk to casually without papering the relationship. Both of those have generated real disclosure problems in litigation.

  7. operator

    And the LLM question — I keep wanting to paste code into ChatGPT to ask for help. What's the real rule?

  8. agent

    Two tiers, two rules.

    Consumer tier (ChatGPT free/Plus, Claude.ai free/Pro, Gemini, etc. without a business agreement): assume anything you paste is a public disclosure. The terms of service allow training. Courts are already treating these as voluntary disclosures — Trinidad v. OpenAI killed a trade-secret claim because the secret was developed in ChatGPT. Heppner found that sharing with a commercial AI waived attorney-client privilege. Samsung banned ChatGPT internally after engineers leaked source code into it.

    Enterprise tier with a signed Data Processing Addendum (DPA) that says no training on inputs, with zero retention or scoped retention: treated as confidential. Same as sending to a contractor under NDA.

    The rule that's safe to use: proprietary code, method detail, and business strategy only ever touch enterprise-tier tools with a DPA. Everything else uses scrubbed inputs — describe the shape of the problem without the secret sauce.

The IP review gate

The reason most disclosure problems happen isn’t that the operator didn’t know the rules. It’s that nobody was the named gatekeeper, so a pitch deck went out, a blog post went up, a prompt got pasted into ChatGPT, and the disclosure happened in the gap between “everyone knows we should be careful” and “one specific person reviews this before it ships.”

The fix is structural: name one IP reviewer and wire them into the workflow as a real checkpoint.Not “the team is careful.” One human (or one agent that escalates to a human) whose signoff is required before any of these artifacts go out:

  • Pitch decks shared outside the cap table.
  • Customer or partner demos where outsiders can observe the system.
  • Blog posts, papers, conference talks, podcast appearances, README updates that describe method detail.
  • Public repos and any code being pushed to a non-private destination.
  • Marketing pages on the company site that describe how the product works.

The reviewer’s job is small: run the pre-publish check (below), call out anything that’s a disclosure of a patentable method, confirm filings are in place or get them in place. Not a legal review — an operator-level check that the order of operations is right.

The companion failure to the disclosure gate is the consumer-LLM gate. Same idea, different surface. Proprietary code, method detail, training data, and business strategy only ever touch enterprise-tier tools with a signed DPA that prohibits training on inputs. Consumer tools get scrubbed inputs — describe the shape of the problem, not the secret sauce. Samsung banned ChatGPT internally after engineers leaked source code into it; the right move is to set the rule before that incident happens, not after.

The pre-publish check — five questions

Before anything about a patentable method goes outside the company, the named reviewer walks these five questions:

  1. Is anyone seeing this who hasn’t signed an NDA? If yes, either get the NDA signed first, or strip the artifact to public-safe content.
  2. Does this read on a claim we’d want to file? If yes, is there a provisional or non-provisional already on file with a priority date before this disclosure?
  3. Do we care about foreign patent rights for this method? If yes, a filing has to be on file beforethe disclosure — there’s no grace period to recover into.
  4. If this is going to a consumer-tier LLM, does it contain proprietary method detail? If yes, scrub or move to an enterprise-tier path. Otherwise, treat the prompt as published.
  5. Has the named IP reviewer signed off? Not the team, not the founder remembering to think about it — the named gatekeeper, in the workflow, with a checkpoint that has to be cleared.

The decision rule that comes out of this: if any answer is “no” or “not sure,” the artifact doesn’t go out until it’s resolved.No exceptions for “it’s just a quick post” or “they’re trustworthy.” Trustworthy people aren’t the failure mode. Procedural gaps are.

Prompt kit

Three prompts for running the IP review gate. The first two run before specific actions; the third is an audit you run once on existing artifacts.

The pre-publish check
I'm about to <publish / post / pitch / demo / share> the following:

<paste the artifact — slide deck, blog post, README, repo, demo plan>

Walk through:
1. Does this artifact describe a method, system, or process that
   could be patentable? If yes, which parts read on what a claim
   would look like?
2. Has anything in this artifact already been disclosed publicly
   before? If yes, when and where? (US grace-period math.)
3. If this is the first public disclosure, what filings should be
   in place BEFORE it goes out to preserve patent rights:
     - US: provisional or non-provisional?
     - Foreign (EPO, China, Japan, Canada)?
4. Is there confidential information in here that doesn't need to
   be public for the artifact to do its job? Suggest scrubs.

Output: a go/no-go recommendation, the filings I should have on
file first, and a redline of anything to strip before publish.
The enterprise-vs-consumer LLM check
I want to paste the following into <ChatGPT / Claude / Gemini /
other LLM> to help with <task>:

<paste the prompt or describe the contents>

Walk through:
1. Does this contain proprietary code, method detail, training
   data, customer data, or business strategy?
2. Is the destination tool consumer-tier (training allowed,
   standard ToS) or enterprise-tier (signed DPA, no-training,
   zero or scoped retention)?
3. If consumer-tier + proprietary content: what's the scrubbed
   version that gets the same answer? (Describe shape, not
   secret sauce.)
4. If proprietary content is unavoidable, name the enterprise
   path that's safe to use.

Output: safe / not safe, the scrubbed version if relevant, and
the right destination.
Audit existing public artifacts for inadvertent disclosure
Read everything I've published in the last 18 months that's
publicly accessible (blog posts, READMEs, public repos, conference
talks, podcast transcripts, pitch decks shared without NDA).

For each artifact:
- What was disclosed about our methods, models, or systems?
- What's the publication date?
- Does any of it read on a claim we'd want to file?

Output a table:
| artifact | date | what was disclosed | grace period status |

Flag anything where the US one-year grace period has closed or is
within 60 days of closing — those are file-now-or-lose-rights items.

Apply this — stand up the IP review gate

One-sitting exercise.Name the reviewer. Wire them into the workflow. Run the pre-publish check on the next artifact. The whole thing takes about an hour and prevents the failure mode that’s unrecoverable.

Stand up the IP review gate

Each step takes 10-20 minutes. Progress saves automatically.

0/5
  1. 01Name the single person who reviews anything before it goes public.One named reviewer. Not "the team." Not "send it around." The IP review gate is one human (or one agent that escalates to a human) whose signoff is required.
  2. 02Wire that reviewer into the workflow as a checkpoint, not a culture norm.If publishing means a PR, the reviewer is a required approver on PRs touching docs/, READMEs, /labs/, marketing pages. If publishing means email, the reviewer is on the from-line or the CC. Mechanical enforcement, not promised future discipline.
  3. 03Inventory the consumer-LLM exposure on your team. Who's using what, with what kind of inputs?Ask the team directly. Most operators are surprised. Either (a) get enterprise-tier accounts with DPAs in place, or (b) write a clear rule about what kinds of content can never be pasted into consumer tools.
  4. 04For any method you'd want to patent, file a provisional BEFORE the next public demo, pitch, or write-up.Costs ~$130 in USPTO fees plus a few hours of drafting (or a couple thousand if you use a patent attorney). Gives you a priority date and 12 months to convert. The cost of skipping this is your foreign patent rights.
  5. 05Run the pre-publish check (from the prompt kit) on the next artifact about to go out the door.Even if you think it's clean. The first time you run it, you'll usually find at least one thing worth scrubbing or filing on first.
Operating tier · what's next

After this lesson