Don't Credit the LLM: Who Actually Owns AI-Assisted Work
Should you credit an LLM in your work? Isaac Su says no, and his reasoning holds up. But there is a line between credit and disclosure that every freelancer and student needs to know.
Whether you should credit an LLM in your work is a question I now see answered badly about twice a week. A GitHub PR opens with "generated with AI, please review carefully." A pitch deck ends with a "made with AI assistance" line nobody asked for. Isaac Su's post Don't credit the LLM makes the case that this habit is not humility. It is accountability leaking out of your work.
I think he is right, and I also think most people will apply his advice in exactly the wrong place.
🔍 The argument, stripped to its frame
Su's observation is that colleagues have started disclosing LLM use when presenting work, and that nobody can explain why. His comparison: a writer who lists the spellchecker they used alongside every piece. His sharper one: a pilot announcing the version number of their avionics software during taxi.
He proposes four motivations behind the reflex:
- Guilt — a felt obligation to credit the tool so you don't seem dishonest.
- Borrowed importance — attaching your work to something impressive.
- Skipped review — you shipped output you hadn't actually read.
- Pre-loaded excuse — a claim of AI use, real or not, that lowers the bar you'll be judged against.
Only the first is well-intentioned. The other three are things you would not say out loud.
Key takeaway: Credit and accountability are two sides of the same coin. A tool cannot be held accountable, so crediting it does not share the load — it just thins yours.
That sentence is the whole post, and it is worth sitting with. When you write "AI helped with this," you have not added information. Your reader still cannot sue the model, fire the model, or ask the model what it meant on line 340. All you did was tell them you are slightly less prepared to answer for it than you were a sentence ago.
⚖️ Why the disclaimer costs you more than you think
Reasons 3 and 4 above are the ones that hurt, because they are the ones your reader assumes even when they aren't true. Look at what the same PR description signals in each direction:
| What you write | What you think it says | What a reviewer hears |
|---|---|---|
| "Generated with AI, please review carefully" | I'm being transparent | I did not review this |
| "AI-assisted, may contain errors" | Setting expectations | The errors are now your problem |
| "Wrote this with Claude's help" | Crediting my tools | The work is not fully mine |
| (nothing) | — | This person stands behind it |
For a freelancer in Colombo bidding against someone in Warsaw or Manila for the same contract, this matters more than it does for a salaried engineer. You are competing on trust with a client who will never meet you. The only thing you are actually selling is that someone answers for the work when it breaks at 2am. A disclaimer at the bottom of your delivery quietly removes the one thing they were buying.
The same logic runs the other way for hiring. If I get a take-home submission with "AI-assisted" in the README, my next question is not did you use AI — I assume everyone did. It is which lines can you explain. That is a much worse interview to walk into than the one where you just said nothing and knew your code.
🇱🇰 The line Su's post doesn't draw: credit vs disclosure
Here is where I'd extend the argument rather than just agree with it. Su is talking about credit — voluntary attribution you offer to make yourself look honest. That is different from disclosure — a declaration someone else has the standing to require of you.
Credit is optional and usually counterproductive. Disclosure is a rule, and you don't get to opt out of it by quoting a blog post.
| Context | Voluntary credit? | Required disclosure? |
|---|---|---|
| Your own blog, side project, portfolio | No. Own it. | No |
| Client deliverable, no clause in contract | No | No — but read the contract first |
| Client contract with an AI clause | No | Yes. Contractual, not ethical |
| University assignment or thesis | No | Depends entirely on your institution's policy |
| Open-source contribution to someone else's repo | No | Whatever CONTRIBUTING.md says |
| Journalism, research, regulated filings | No | Usually yes, and it will be specific |
⚠️ If you are a student at a Sri Lankan university, do not treat "don't credit the LLM" as permission. Academic integrity policies here vary by faculty and are being rewritten right now. Go read the current one for your programme before you decide. Su's argument is about presenting professional work, not about coursework rules.
The distinction is clean once you see it: credit is about sharing glory, disclosure is about following a rule you agreed to. Su is arguing against the first. Nothing in his post gets you out of the second.
🛠️ The review discipline that makes the disclaimer unnecessary
If reason 3 — shipping unread output — is the real one, the fix is not a warning label. It is reading the thing. What I actually do before anything leaves my hands:
- Diff the draft against my edit. If the diff is nearly empty, I haven't done my job. Our text diff checker is genuinely useful for this: paste the raw generation and your final version, and see how much of it you actually touched.
- Verify every number, name, and date. Models produce confident specifics that don't exist. One fabricated statistic in a client report costs more than the entire time saved.
- Explain each non-obvious block out loud. If I can't say why a function is written that way, it is not mine yet and I rewrite it.
- Delete anything I kept only because it sounded good. This catches most of it.
Run that loop and the impulse to disclaim disappears on its own, because there is nothing left to hedge about. The disclaimer was never a transparency measure. It was a symptom.
💡 What this means for you
Su's framing is a useful test to carry around: if the tool cannot be held accountable, don't put its name on the work. Your compiler isn't in your commit message. Your IDE isn't in your CV. Neither is your model.
Practically:
- Strip the disclaimers from work you stand behind. If you can't stand behind it, that's the actual problem — fix that instead.
- Read your contracts and your course policies. Required disclosure is a separate obligation and it overrides all of this.
- Never claim AI use as an excuse. It reads as an excuse and it is remembered as one.
- Take the failure too. The reason full credit feels good is that full responsibility comes attached. You don't get one without the other, and the trade has always been worth it.
The people who build a reputation over the next few years will not be the ones who used the best models. Everyone will have the same models. They will be the ones who still answer for the output.
Original source
Don't credit the LLM