General
What is the IDK License?
The IDK License (I Don't Know License) is a family of four license variants — from maximally permissive to strong copyleft — designed for code and other works created with AI assistance. It addresses the fundamental uncertainty: when AI tools helped create your code, you may not know which parts are copyrightable, who holds the copyright, or whether the code implements someone else's patents or trade secrets.
Instead of pretending to grant rights you may not have, the IDK License honestly acknowledges this uncertainty. Each variant uses layered legal strategies — each only to the extent the author legally can — including fallback license grants and patent non-assertion covenants.
Why not just use MIT or Apache-2.0?
Traditional licenses like MIT and Apache-2.0 implicitly assume the licensor knows they hold copyright and have the right to grant a license. The standard MIT license begins "Permission is hereby granted..." — but what if you don't actually have the legal standing to grant that permission?
When AI tools contribute to your code, that assumption breaks down. Copyright law in most jurisdictions is unsettled on AI-generated output (see Thaler v. Perlmutter, US Copyright Office guidance, etc.). The IDK License explicitly handles this uncertainty with fallback strategies.
If you're confident that your work is entirely human-authored and you hold full copyright, MIT or Apache-2.0 work fine. The IDK License is for when you honestly don't know.
Is this license OSI-approved?
No. None of the IDK License variants are (yet) submitted to or approved by the Open Source Initiative. However, each variant simultaneously grants the work under well-known fallback licenses that are OSI-approved. For example, the base IDK License includes MIT, Apache-2.0, BSD-2-Clause, BSD-3-Clause, 0BSD, ISC, Boost-1.0, zlib, and MIT-0 among its 15 fallbacks. Recipients can treat the work as licensed under whichever applicable fallback they prefer.
Is this license GPL-compatible?
Three of the four IDK License variants are GPL-compatible:
- IDK and IDK-Attribution include a one-way copyleft compatibility grant (IDK Section 2.5; IDK-Attribution Sections 2.5–2.6) that explicitly permits incorporating their code into projects under GPL-2.0, GPL-3.0, LGPL-2.1, LGPL-3.0, AGPL-3.0, MPL-2.0, EPL-2.0, and EUPL-1.2.
- IDK-Weakleft is copyleft and includes fallback multi-license grants under AGPL-3.0, GPL-3.0, LGPL-3.0, MPL-2.0, EPL-2.0, EUPL-1.2, and IDK-Strongleft, plus explicit compatibility clauses in Section 17.
- IDK-Strongleft is strong copyleft with only AGPL-3.0 as its fallback. IDK-Strongleft code cannot be relicensed under GPL-3.0 (AGPL is more restrictive than GPL).
Can I use this for non-code works (documentation, images, data)?
Yes. While the IDK License family is designed with software in mind, each variant applies to any copyrightable (or potentially copyrightable) work. Section 13 defines "Work" broadly. The base IDK and IDK-Attribution variants include CC0-1.0 and CC-BY-4.0 among their fallbacks, which are commonly used for non-software works.
Is this a joke license?
No. Despite the informal name, the IDK License family is a serious attempt to address a real legal problem. Each variant uses layered strategies — including fallback license grants and patent non-assertion covenants — based on established legal mechanisms used by CC0, Apache-2.0, GPL, and other respected licenses.
That said, it was drafted with AI assistance and has not been reviewed by a licensed attorney. It comes with no warranty. You should treat it as a good-faith effort, not as a proven legal instrument.
License Variants
What are the different IDK license variants?
The IDK License family has four base variants, from most permissive to most restrictive:
- IDK — Maximally permissive (CC0-like). Public domain dedication + 15 permissive fallback licenses + copyleft compatibility grant. No conditions at all.
- IDK-Attribution — Permissive with attribution (MIT-like). 9 attribution-requiring fallback licenses + copyleft compatibility grant. Requires preserving copyright and license notices.
- IDK-Weakleft — Copyleft (GPL-like). 7 copyleft fallback licenses (AGPL, GPL, LGPL, MPL, EPL, EUPL, IDK-Strongleft). Derivatives must be open source under the same terms.
- IDK-Strongleft — Strong copyleft (AGPL-like). AGPL-3.0 fallback license. Also requires source disclosure for network deployments.
Each has a WTF ("With These Footnotes") wrapper for projects with third-party dependencies. See the comparison page for a full feature matrix.
What's the difference between IDK-Strongleft and IDK-Weakleft?
Both are copyleft licenses that require derivative works to be distributed under the same terms with source code. The key difference is the network interaction clause:
- IDK-Weakleft (GPL-like): Only triggers when you distribute the work. Running it as a SaaS/web service without distributing copies does not require source disclosure.
- IDK-Strongleft (AGPL-like): Also triggers when users interact over a network. Running it as a web service requires offering source code to those users.
How does copyleft work when AI authorship is uncertain?
Traditional copyleft relies on copyright to enforce share-alike obligations. If AI-generated code isn't copyrightable, copyright-based copyleft may not apply. The IDK copyleft variants (Weakleft and Strongleft) address this with a contractual backstop (Section 5.4):
- The copyleft obligations operate as both copyright conditions and independent contractual covenants.
- Even if copyright enforcement fails for some portion, the contractual obligation to share source remains.
- This is a novel approach — no existing license does this.
Can I mix code from different IDK variants?
Compatibility flows in one direction, from more permissive to more restrictive:
- IDK code can go into any other variant (IDK-Attribution, Weakleft, or Strongleft)
- IDK-Attribution code can go into Weakleft or Strongleft projects
- IDK-Weakleft code can go into IDK-Strongleft projects (but not the reverse)
- IDK-Strongleft code cannot be incorporated into any less-restrictive variant
IDK, IDK-Attribution, and IDK-Weakleft are also compatible with GPL-3.0. IDK-Strongleft is compatible only with AGPL-3.0 (its sole fallback). See the comparison page for the full compatibility matrix.
Legal
What does "to the extent the author legally can" mean?
This qualifier appears throughout the IDK License. It means: every grant, dedication, and waiver in the license only takes effect if and to the extent the author actually has the legal right to make it. If the author has no copyright in some portion of the work (because it's not copyrightable, or because it was generated by AI, or for any other reason), the license is simply ineffective for that portion — it doesn't claim to grant what the author doesn't have.
This is a fundamental principle of law (nemo dat quod non habet — you can't give what you don't have). The IDK License makes this explicit rather than leaving it implicit.
What if AI-generated code isn't copyrightable?
If a portion of the work is not copyrightable (e.g., because it was purely AI-generated and your jurisdiction doesn't recognize AI authorship), then no copyright license is needed for that portion — it's already free for anyone to use. The IDK License's grants simply have no effect for that portion, which is fine because no permission is needed.
The patent provisions (Section 3) still apply regardless of copyrightability, since patent rights are independent of copyright.
Does this license protect me from patent claims?
From the author's patents: Yes, to the extent the author holds patent claims. Section 3 includes a patent non-assertion covenant and an explicit patent grant, with defensive termination if the recipient sues first.
From third-party patents: No. Section 3.5 explicitly warns that AI-generated code may unknowingly implement patented methods belonging to third parties. The author cannot protect you from claims they don't know about. This is a risk inherent in AI-assisted development, and the IDK License is honest about it.
What about trade secrets?
Section 3.6 addresses trade secret risk. AI models are trained on large datasets that may include proprietary or trade-secret-protected code. If an AI tool reproduces such code in its output, the resulting work could contain trade secrets without anyone's knowledge. The author cannot warrant that the work is free from such contamination.
Unlike patent infringement (which is strict liability), trade secret misappropriation typically requires knowledge or reason to know. If you receive IDK-licensed code in good faith, you may have an "innocent recipient" defense — but this depends on your jurisdiction and the specific facts.
Has a lawyer reviewed this license?
No. The IDK License was drafted with AI assistance and has not been reviewed by a licensed attorney in any jurisdiction. This is disclosed prominently at the top of the license text. The irony is intentional: a license about not knowing the legal status of AI-generated work is itself AI-generated and of uncertain legal status.
If you are considering using this license for important projects, you are strongly encouraged to have it reviewed by an attorney familiar with intellectual property law in your jurisdiction.
Can I revoke the IDK License after applying it?
No. Section 9 makes the license irrevocable. Once you release a work under the IDK License, you cannot revoke the grants for copies already distributed. The license also includes an estoppel provision: even if some legal theory would allow revocation, you are estopped from asserting it against anyone who relied on the license in good faith.
What happens if part of the license is found invalid?
Section 10 provides enhanced severability. If any provision is found invalid or unenforceable, it is severed and the rest continues in full force. This includes blue-pencil and equitable-reduction rules to preserve maximum intent. Importantly, invalidation of one fallback license does not affect the others (the IDK License has 15 fallbacks; other variants have fewer).
This extends to the content of the Work itself (Section 10.7): if any portion of the work turns out to be uncopyrightable, not owned by the author, or otherwise unlicensable, the license continues to apply to all remaining portions. The license treats each severable portion of the work independently, so a problem with one part cannot undermine the grants on the rest.
Technical
What SPDX identifier should I use?
Each variant has its own SPDX identifier following the LicenseRef-
convention for non-registered licenses:
LicenseRef-IDK-0.0.1LicenseRef-IDK-Attribution-0.0.1LicenseRef-IDK-Weakleft-0.0.1LicenseRef-IDK-Strongleft-0.0.1
WTF wrappers:
LicenseRef-IDK-WTF-0.0.1LicenseRef-IDK-Attribution-WTF-0.0.1LicenseRef-IDK-Weakleft-WTF-0.0.1LicenseRef-IDK-Strongleft-WTF-0.0.1
// SPDX-License-Identifier: LicenseRef-IDK-0.0.1
How do I apply this license to my project?
For a project with no third-party dependencies:
- Copy
LICENSE-IDKto your project root. - Add
// SPDX-License-Identifier: LicenseRef-IDK-0.0.1to your source files. - Mention the license in your README.
See the IDK License page for full quick-start instructions, or Appendix B of the license text for detailed guidance including REUSE compliance.
What is the dependency scanning agent?
The IDK-WTF License includes an AI agent prompt (AGENT-PROMPT.md)
that can scan your project for third-party dependencies and automatically
populate the Dependency Schedule. It supports 17+ package ecosystems
(npm, pip, Cargo, Go, Maven, etc.) and handles:
- Parsing lockfiles and manifests
- License detection from metadata and file contents
- SPDX identifier normalization
- Vendored and bundled code detection
- Generating the correctly-named IDK-WTF output file
There is also an update agent (AGENT-PROMPT-UPDATE.md) that
can incrementally update an existing IDK-WTF file when dependencies change,
computing diffs and preserving manual annotations.
What's the file naming convention for IDK-WTF?
IDK-WTF files are named to reflect the third-party licenses they contain:
IDK-WTF--<licenses>.md
Examples:
IDK-WTF--MIT.md— one dependency under MITIDK-WTF--Apache-2.0-OR-MIT+BSD-3-Clause.md— one dual-licensed Apache/MIT + one BSD-3-ClauseIDK-WTF--AGPL-3.0-only-OR-GPL-3.0-only+Apache-2.0-OR-MIT+BSD-2-Clause.md
Licenses within a choice group use -OR-, separate works use +,
conjunctive licenses use -AND-, and exceptions use -WITH-.
See Part IV of the IDK-WTF License for full rules.
Is the IDK License compatible with REUSE?
Yes. Appendix B of the license text includes specific REUSE compliance
guidance. You can place the license text in a LICENSES/ directory
as LICENSES/LicenseRef-IDK-0.0.1.txt and use standard SPDX
headers in your source files.
WTF Wrappers (With These Footnotes)
When should I use a WTF variant instead of the base license?
Use a WTF variant whenever your project includes third-party code — libraries, frameworks, vendored files, submodules, copied snippets, or any code you didn't write (or have AI write for you). This is most real-world projects.
Use the base variant only for projects where every line of code is either yours or AI-generated with no external dependencies.
Each base license has its own WTF wrapper: IDK-WTF, IDK-Attribution-WTF, IDK-Weakleft-WTF, and IDK-Strongleft-WTF.
Does the IDK License apply to my dependencies?
No. The IDK License (via IDK-WTF Part II) applies only to your original work. Third-party code keeps its own license. The Dependency Schedule in Part III helps recipients identify which license applies to which component, but each third-party license governs independently.
What if a dependency uses a copyleft license (GPL, AGPL, etc.)?
Part V of the IDK-WTF License addresses copyleft interaction. The copyleft terms of the third-party work govern that work and any derivative work incorporating it, as required by the copyleft license. The IDK License's Section 2.5 compatibility grant means your original work can be incorporated into copyleft projects without triggering compatibility issues.
However, you (the author) must still comply with the copyleft license's requirements for the third-party code (source distribution, license preservation, etc.).
Is the Dependency Schedule guaranteed to be accurate?
No. The license explicitly disclaims any warranty of completeness or accuracy (Part II and Part V.4). If the schedule was generated by the scanning agent, its accuracy depends on the agent's ability to detect dependencies. Manual review is recommended. If it was populated manually, it's only as accurate as the person who filled it in.