GitHub trending is now a shadow market for AI-service abuse tools

What happened

Three repositories surfaced on GitHub trending this week that, viewed together, tell a single story. `yynxxxxx/Codex-5.5-codex-instruct-5.5` cleared 1,332 signals — a repo whose name alone claims a model version OpenAI has never shipped and a lineage (Codex) that OpenAI officially retired in 2023 before resurrecting the brand for its coding agent. `akihitohyh/chatgpt-register-sub2api` sits at 181 — the URL slug is the pitch: automate ChatGPT account registration, then expose the session as an OpenAI-compatible API. And `LinXiaoTao/FuckClaude` posted 340 signals with a name that isn't asking for subtlety.

None of these are hobby projects mislabeled. They're a category. The unifying feature isn't a technology; it's a business model — extract paid AI capacity from a vendor's consumer subscription and resell or reuse it through an API surface the vendor didn't sanction. The Codex-branded repo trades on model-name confusion to attract developers searching for something official. The `sub2api` repo is transparent about what it does: convert a subscription into an API. The Claude repo names its target in the URL.

All three trended on GitHub in the same week. That's not a lull in real releases — the Linux kernel and freeCodeCamp are still on the same trending page. It's a signal that a specific kind of tooling has found product-market fit inside the trending algorithm itself.

Why it matters

Trending on GitHub used to be a rough proxy for what practitioners cared about. New database, new framework, new build tool. That signal is degrading, and the degradation isn't random. The three repos above are optimized for GitHub's discovery surface the same way affiliate spam is optimized for Google: names carry the query, the README carries the CTA, and the star count carries the trust.

Compare the naming conventions. `Codex-5.5-codex-instruct-5.5` reads like a Hugging Face model card someone tried to launder into a repo name — the doubled version tag, the `-instruct` suffix borrowed from open-weights convention, the '5.5' that splits the difference between GPT-5 and a hypothetical mid-cycle bump. If you're a developer searching for 'codex 5.5' because you saw the term in a Discord, this repo is the first result. `chatgpt-register-sub2api` uses `sub2api` as if it's a known noun — and increasingly, it is. Search that string and you'll find dozens of forks: `claude-sub2api`, `gemini-sub2api`, `poe-sub2api`. It's a category name in a market GitHub isn't officially hosting but is de facto indexing. `FuckClaude` is a discovery hack of its own kind: vendor-name-plus-expletive filters out corporate search results and puts you directly in front of the workaround crowd.

The community reaction to this pattern is bifurcated in a predictable way. Practitioners who've been burned once — by a supply-chain attack, a leaked key, a Terms of Service ban that killed a side project — see these as obvious hazards. Everyone else sees free tokens. The gap between those two groups is where the risk lives, and it's growing because the tooling is getting more polished. Early sub2api repos were shell scripts held together with `curl` and hope. The current generation ships proper OpenAI-compatible endpoints, streaming support, rate-limit backoff, and Dockerfiles. They look like real infrastructure because, functionally, they are — just pointed at accounts the operator doesn't own or licenses that don't cover API access.

There's also a quieter risk: credential theft dressed as bypass tooling. A repo that promises to 'convert your ChatGPT sub to an API' has to accept your session cookie or your login credentials somewhere. Nothing in the trending signal tells you whether that data stays on your machine or gets shipped to a Telegram bot the moment you run the container. The `FuckClaude` naming pattern is worse for this — the audience self-selects for people willing to ignore red flags, which is exactly the audience malware authors want.

What this means for your stack

If you run an engineering org, treat trending-derived AI tooling the way you already treat trending npm packages: with a bill-of-materials review before anything touches a machine with credentials on it. The threat model isn't 'my developer downloaded a bypass tool' — it's 'my developer downloaded a bypass tool that also reads `~/.aws/credentials` on startup because the operator knew who the audience would be.' Add the `sub2api` string, the vendor-name-plus-expletive pattern, and any repo claiming a model version your vendor hasn't announced to your internal blocklists or Semgrep rules.

If you're a solo dev tempted to save on API costs, the honest calculation is worse than it looks. TOS violations get accounts banned, and the ban usually catches the sub2api operator's whole customer base at once — you'll lose access mid-project, without warning, and without recourse. Legitimate alternatives now exist that didn't a year ago: OpenRouter's pay-per-token pooling, Anthropic's prompt caching (up to 90% cost reduction on repeated context), and the increasingly usable open-weight models on Groq and Together will beat any bypass tool on total-cost-of-ownership once you price in the switching cost of a ban.

For GitHub itself, the trending page is becoming an SEO surface for a gray market it doesn't want to host. That's a moderation problem the platform has historically been slow on — DMCA and malware take years to enforce, and 'facilitates TOS violation of a third party's paid service' is a much fuzzier standard than either. Expect vendors to start filing takedowns directly. Anthropic and OpenAI both have the legal muscle and the motive.

Looking ahead

The trending page as a signal is worth keeping an eye on, not for what's on it but for what the composition tells you about incentive gradients in the AI market. When bypass tooling out-trends real releases in a given week, that's a datapoint about pricing pressure — devs cannot afford, or believe they cannot afford, the sanctioned API at current rates. The vendors will respond, eventually, with lower per-token prices or better free tiers. Until then, the sub2api category will keep filling the gap, and the trending page will keep advertising it.