"Humble Bundle just killed most of their security through vibecoding"

Humble Bundle’s recent website overhaul has drawn sharp criticism for stripping away basic account security features, breaking purchase histories, and introducing dark-pattern payment flows such as one‑click charges without clear confirmation. Commenters speculate that the rewrite was done cheaply, possibly with heavy reliance on AI “vibecoding” and outsourced or reduced engineering staff, leading to a product that appears less secure and less usable than the old site. The incident is being cited as a cautionary example of how management pressure to cut costs and overtrust in AI tools can amplify incompetence and ship dangerously low‑quality software.

Humble Bundle Redesign & Functional Regressions

  • New site widely seen as a major downgrade: “generic gaming site” look, heavy gradients, blurry hero images, slow hover menus, broken or empty sections.
  • Many report login problems (Google/email failing, repeated CAPTCHAs, HTTP 431 errors likely from cookie bloat).
  • Purchase history and libraries appear incomplete, with “still migrating” messages; pagination shows only a few items per page, making large libraries hard to use.
  • Some legacy content seems missing or corrupted; support responses reported as “tough luck.”

Security & Legal Concerns

  • 2FA removed: password-only logins now; seen as a huge regression. A minority are happy to lose mandatory 2FA friction.
  • Password change flow reportedly allows changing without confirming old password or re-entering the new one.
  • At least one-click payments triggered charges from a vague “Get your games” button; commenters note this likely violates EU consumer rules requiring explicit purchase confirmation, especially in places like Germany.
  • Account deletion is difficult: settings lack a clear option; users must use a separate DSAR link and follow a multi-step process.

AI, Vibecoding, and Responsibility

  • Many assume the site was built via “vibecoding” (LLM-driven development with weak understanding/review), but others stress there’s no direct proof; insecure software predates AI.
  • Debate centers on whether “AI produces bad software” or “people produce bad software using AI”; analogies used include saws, cars, guns, and bulldozers.
  • Some argue LLMs massively amplify incompetence by enabling inexperienced people to ship large amounts of unreviewed code.
  • Others insist the core problem is management: layoffs, outsourcing, cost-cutting, and overbelief in AI as a dev replacement.

Rewrites, Testing, and Process

  • Many see this as a textbook “big bang rewrite” failure: regressions in security, UX, and functionality, with poor migration and no visible rollback plan.
  • Several note that even minimal manual testing or comparing old vs new flows (e.g., 2FA) should have caught obvious issues.
  • Commenters link this to broader industry patterns: under-valuing expertise, rewriting working systems “because AI makes it cheap,” and ignoring long-term maintainability.

User Sentiment and Reactions

  • Long-time customers report cancelling subscriptions and deleting accounts out of distrust.
  • Some express nostalgia for the pre-acquisition era and perceive the brand as “on life support.”
  • General consensus: release is “extremely unprofessional” and a warning sign of where AI-assisted, cost-driven rewrites can lead.