The Case of the Robot Web Thief

There’s been buzz recently about how a nascent web and mobile astronomy app, Dark Hours, got wildly incorrectly rejected from Apple’s App Store. Then maybe it turned out that the rejection wasn’t so wild because Apple may well have been looking at the astrology app the developer first submitted under the name “Asterly”. Then it turned out that the web version of Dark Hours was strikingly similar—although UI-wise, hardly identical—to an existing web app called DarkHours (camelcase rather than two words). It was so similar, in fact, that they both shared the same weird bug. All this led to Daring Fireball’s first-ever retraction and considerable opprobrium directed at Terry Godier, developer of Dark Hours (two words rather than camelcase).

Godier was up front about creating his web app with Claude, writing in his apology,

I was careless in relying on AI to generate the project without doing the work to understand whether it closely resembled an existing project. That’s on me, and I am responsible for what I published.

And, yes, sure, but I think this raises fascinating questions about letting an AI do most or all of your work that run beyond these specifics.

• It’s probable that Godier would have found the existing DarkHours if he did a non-AI web search for something like, you know, astronomy app dark hours. But DarkHours’s source code is available on GitHub, so it’s likely in Claude’s training data. If the names weren’t the same, this could still easily have happened, and it might have taken much longer for anyone, including Godier, to discover it.
• And camelcase-not-two-words DarkHours was also co-written with Claude. The supposed “smoking gun” of the identical bug could have existed not only with two differently-named apps, but with DarkHours not being on GitHub: Claude very likely would code a similar function in a similar way, making similar mistakes. (In fact, presuming that bug has been fixed in the publicly available source code, that’s almost certainly what happened.)

“Check how close both your idea and your branding is to existing apps” is always sound advice, but using AI makes this considerably more fraught. You may have no idea if Claude’s effectively pulled a chunk of a GPL-licensed library into your proprietary code, or replicated an aspect of an existing app in a way that doesn’t technically violate licensing or copyright but opens you up to a patent lawsuit.

It’s easy to write off Godier’s misadventures here as a one-off mess he got himself into, but the implications of said misadventure are far-reaching. Lawyers at tech firms should, at the very least, begin insisting on very strict review and vetting procedures for any AI-written code, and indie developers who’ve been relying on agentic AI need to do the same. Just checking that Claude’s code reads well and runs correctly isn’t enough.