HN Debrief

Ask HN: Is "no source code was copied" still a sufficient copyright defense?

  • AI
  • Copyright
  • Programming
  • Open Source
  • Regulation

The post used the recent Corgi versus Papermark flap as a prompt for a broader question: if an LLM helps you recreate a competing app without directly copying source files, does that still count as infringement. The comments landed on a practical answer, not a clean doctrine. "No source code was copied" can help in a classic clean-room defense, but it is nowhere near a complete shield if the copied product includes protected text, images, design assets, or a UI that is protected under local law or design-patent-like regimes. Several people also pointed out that software copyright never covered ideas, workflows, protocols, APIs, or business logic in the first place, so the dispute quickly spills beyond source code into trade dress, design rights, contract terms, and unfair competition.

If you are shipping AI-assisted clones, do not rely on "the code is different" as your legal position. Audit copied text, graphics, UI assets, license obligations, jurisdiction, and contract terms up front, because litigation risk now comes from the whole product, not just the repo diff.

Discussion mood

Skeptical of broad software copyright claims and doubtful that "different code" settles anything. The mood was cynical about enforcement, impatient with IP maximalism, and newly aware that AI makes copying cheaper while also making old legal boundaries around code, UI, and authorship much messier.

Key insights

  1. 01

    US registration determines real leverage

    Registering software with the US Copyright Office is what turns abstract ownership into something you can actually enforce. Without registration, a plaintiff is mostly reduced to making threats, because monetary and statutory damages are off the table and even getting into court becomes harder. That makes early registration a practical deterrent against bad-faith copiers, not just paperwork.

    If your product has any real litigation value, register it before a dispute starts. Treat registration as part of release and compliance ops, not as something to think about after copying is discovered.

      Attribution:
    • wahern #1
  2. 02

    Clean room still matters when done for real

    The old BIOS and Unix reimplementation cases still frame the modern answer. If a team works from specifications and avoids access to the original code, copyright law generally allows a fresh implementation because it protects expression, not ideas or functionality. That defense falls apart fast once people, tools, or prompts have seen the original implementation and start echoing its text or assets.

    If you want a defensible reimplementation, separate spec work from implementation and document that separation. Do not assume AI tooling preserves that boundary for you.

      Attribution:
    • dlcarrier #1
    • kube-system #1
    • throwaway81523 #1
  3. 03

    GUI protection changes by jurisdiction

    The hardest mistake in this conversation is assuming US rules travel cleanly. Commenters noted that Europe and the UK can protect the visual result of software more directly, including GUIs through copyright or registered design, even where US doctrine is narrower. A clone that looks legally safe in one market can create exposure in another.

    Do not let a US-only legal read drive global product decisions. If you operate in Europe or the UK, review interface copying under local design-rights rules before launch.

      Attribution:
    • dijksterhuis #1
    • echoangle #1
    • victorbjorklund #1
  4. 04

    LLMs do not wash away infringement

    Using Claude or another model does not break the chain of copying if the output pulls from protected code or assets. The sharper point was that AI adds a new access problem. If a model was trained on the original project or can search to retrieve it, a plaintiff may argue that the output was not independently created at all. That keeps license violations and substantial-similarity claims alive even when no developer manually copied files.

    For AI-generated product code, keep records of tools, prompts, and any retrieval features used. You will need provenance if you ever have to show the output was independently created.

      Attribution:
    • kmeisthax #1
  5. 05

    Contracts can bite even without copyright

    A few comments usefully pulled the issue out of copyright entirely. If a competitor accessed a product under clickwrap or terms that ban copying, scraping, or reimplementation, breach of contract may be the cleaner claim. That matters because companies often focus on whether code or UI is copyrightable and miss the simpler theory sitting in their own terms of service.

    Review your terms before exposing demos, data rooms, or internal tools to outsiders. Contract restrictions may be easier to enforce than trying to prove a novel interface is protected expression.

      Attribution:
    • dataviz1000 #1
    • echoangle #1

Against the grain

  1. 01

    Most claims are opportunistic, not principled

    This view rejects the idea that software suddenly discovered a coherent copyright ethic. The argument is that people invoke copyright when it hurts a company they dislike and dismiss it when it blocks behavior they want, like abandonware or remixing. That makes a lot of the outrage around app cloning look less like settled legal principle and more like selective moralizing.

    Be careful about reading social consensus as legal consensus. If you need a policy for your company, write one that survives cases where the copier is someone you like.

      Attribution:
    • arjie #1
    • jerf #1
  2. 02

    AI may collapse software IP economics

    A more radical camp argued that cheap AI replication exposes software copyright as a weak foundation for value capture. In that framing, information is not scarce, so businesses built on controlling copies will keep losing ground, while service, support, data, and execution become the only durable models. The point is not that copyright disappears tomorrow. It is that betting your moat on it gets worse every year.

    Stress-test your product strategy as if legal exclusivity will weaken, not strengthen. Invest in moats that survive commoditized copying, especially proprietary data, customer relationships, and operational excellence.

      Attribution:
    • otekengineering #1
    • glimshe #1
    • thomastjeffery #1

In plain english

clickwrap ↩
A contract users accept by clicking a button or checkbox, such as agreeing to terms of service before using a product.
GUI ↩
Graphical user interface, the visual screens, buttons, menus, and layout people use to interact with software.
LLM ↩
Large language model, a type of AI system trained on massive text data to generate and analyze language.
registered design ↩
A legal right used in places such as the UK and EU to protect the visual appearance of a product, including interface elements in some cases.
statutory damages ↩
Damages set by law that a court can award without the plaintiff proving the exact amount of financial harm.
trade dress ↩
A form of intellectual property law that can protect the overall commercial look and feel of a product or brand when it identifies the source.

Reference links

Background on the triggering dispute

Copyright and software case law

Jurisdiction and design-right references

Related concepts and historical references