Three of GitHub's top repos today are malware bait. The trending algorithm is broken.

What happened

On the morning of May 25, 2026, three repositories sitting near the top of GitHub's trending signal share more than just timing. `larajuniorlara/Claude-Design-Studio` ("Claude Design AI 2026: Ultimate UI/UX Generator & Plugin Suite – Free Download"), `sofian160616/Delta-Inject-Workstation` ("Delta Executor 2026 ⚡ Ultimate Roblox PC Script Hub - Free Download"), and `bollahouse/office-2024-pro-integration-suite` ("Microsoft Office 2026 Premium Free Download – Full Suite Installer 🚀") all hit engagement scores within one point of each other (422, 421, 421). All three reuse the same README template: rocket emoji, "2026," "Ultimate," "Free Download," a branded name that hijacks search intent for a real product.

None of the three repos contain meaningful code — they exist to drive clicks to an external installer, which security researchers have repeatedly documented as wrappers around info-stealers like Lumma, RedLine, and Vidar. The pattern is not new. The Roblox "Delta Executor" lure in particular has been a known malware delivery vehicle since at least 2023, repeatedly reskinned and re-uploaded. What is new is that this generation of the campaign has expanded from gaming cheats to brand-jacking Claude (Anthropic's coding assistant) and Microsoft Office in the same trending window.

The accounts themselves give the game away: low-history users, no other meaningful contributions, repo names that don't match their actual contents (`office-2024-pro-integration-suite` rendered as "Microsoft Office 2026"). This is the SEO-spam playbook ported to a developer platform — and it's working.

Why it matters

GitHub Trending is treated by a surprising amount of the developer ecosystem as a curated signal. RSS feeds, newsletters, "awesome" lists, and even some IDE plugins pull from trending or from the underlying star-velocity signal that trending derives from. When three malware lures can sit at score 422 simultaneously without intervention, every downstream consumer of that signal is, by transitive trust, recommending malware delivery to their readers.

The deeper problem is structural. GitHub's trending algorithm rewards star velocity over a short window, which is exactly the metric easiest to manipulate with bought stars (a thriving Fiverr market) or coordinated low-quality accounts. There is no quality gate — no signing requirement, no minimum account age for trending eligibility, no language-model classifier flagging "Free Download Installer" READMEs that contain no source code. Compare this to npm, which at least has `npm audit` and a public typosquat-detection effort, or to the Chrome Web Store, which does manual review on suspect listings. GitHub's trending surface, in 2026, has essentially zero adversarial review.

The community has been complaining for years. A thread on Hacker News in 2024 documented the same Roblox-executor pattern with screenshots. Bleeping Computer ran a piece on Lumma stealer campaigns distributing via GitHub releases in late 2024. Microsoft, which owns both GitHub and the brand being impersonated in one of today's lures, has shipped no public anti-spam improvement to trending in that window. When the platform owner can't or won't protect its own trademark inside its own product, the signal that something is structurally wrong is hard to miss.

There's also a second-order effect that matters for the AI tooling space specifically. "Claude Design AI 2026" is the first instance I've seen in this campaign of an Anthropic product being brand-jacked at scale on GitHub. Expect every named LLM product — Claude, Gemini, GPT, Cursor, Windsurf — to be impersonated by similar lures in the next quarter as the campaign generalizes. If you're a dev rel person at one of these companies, you should be subscribed to a search alert for your product name on GitHub right now.

What this means for your stack

If you or your team uses GitHub Trending as a discovery mechanism, stop. Replace it with one of: curated lists with human editors (Hacker News front page, lobste.rs, this site), GitHub's own search with `stars:>500 created:<2024` filters that exclude the velocity-gaming window, or trust-graph signals (who you follow stars what). Star count alone, especially recent star count, is now an actively adversarial metric.

If you're building tooling that ingests GitHub data — dependency scanners, supply-chain risk products, "what's trending in X" newsletters — add a content classifier. The README signals are extremely cheap to detect: "Free Download" + a year in the future + a brand name + no source files is a 99%-accurate heuristic. You can implement this in a 20-line Python script with the GitHub API and a regex. The fact that GitHub hasn't is the story.

If you're a security team auditing what your devs are running locally, add GitHub Releases binaries to your EDR coverage. The actual malware payload in these campaigns is typically a `Setup.exe` or `Installer.exe` distributed via the Releases tab, which is not subject to the same scanning as the repo source itself. Releases binaries are a known blindspot in most corporate AppSec coverage.

Looking ahead

The interesting question is not whether GitHub will fix trending — they've had years and haven't — but whether a credible alternative discovery layer emerges. The opportunity for a "signed, source-required, human-reviewed" trending feed is real, and the addressable market is every developer who has ever clicked through to a Lumma stealer thinking they were getting a free copy of Office. Whoever ships a curated alternative with a real spam team — even as a paid service — will eat GitHub's lunch on the one part of the platform that actually drives discovery. Until then, treat the trending page like the comments section of a 2010 download site: there's signal, but you have to assume malice by default.