Skip to content
Tech

Why do my AI-generated tests pass on the bugs they were meant to catch?

83

Szansa

AI coding assistants generate unit tests by reading the implementation, so assertions reflect what the code currently does rather than what it should do. A test written against a buggy function learns the bug as a fixture, not a violation. Developers shipping with 90% AI-generated coverage see mutation scores stay flat and real regressions slip through unchanged. The core issue is the oracle problem: there is no ground truth for correct behavior unless you supply it, and AI assistants have no access to the requirements or intent behind the code. Research from July 2025 confirms that LLM-generated tests frequently pass on the exact buggy code they were written against, failing only when the bug is fixed.

Dlaczego to ważne

A test suite that passes on broken code is worse than no test suite, because it turns the primary quality signal into a false guarantee.

Jak oceniam szansę

Wskaźnik Szansy to mój własny odczyt, a nie pomiar: jak bardzo boli, jak często daje się we znaki i jak niewiele istnieje dziś, by to rozwiązać. Wyższy wynik oznacza, że uważam problem za bardziej warty zbudowania.

Dotkliwość8/10

Ile bólu sprawia, gdy się pojawia.

Częstotliwość9/10

Jak często ludzie faktycznie na to trafiają.

Białe znaki8/10

Jak mało dobrych narzędzi istnieje dla tego dziś.

Więcej problemów wartych rozwiązania

Tech

Dlaczego oprogramowanie, od którego jesteśmy najbardziej zależni, jest najgorsze w użyciu?

Tech

Data ownership is still effectively zero for most people because of a few structural problems that reinforce each other: **The terms-of-service trap** Every platform you use grants itself a perpetual, irrevocable, sublicensable license to your content. You "agreed" to this to get access. There is no meaningful negotiation -- it is take-it-or-leave-it, and the network effects make leaving it impractical. **Data is not legally property** In most jurisdictions, raw data has no clear property status. Copyright protects creative expression, not facts or behavioral signals. There is no law that says your location history, click patterns, or biometric data is *yours* the way a car or a house is yours. Without property rights, there is no theft, no trespass, no remedy. **GDPR gave access rights, not ownership** Europe's GDPR is the strongest regime in the world and it gives you the right to *access*, *correct*, and *delete* your data. That is closer to a tenant's rights than to ownership. You cannot sell it, license it, or sue if someone profits from it without your consent. **The commercial model depends on the asymmetry** Surveillance-based advertising -- which funds most of the free internet -- only works if users do not extract economic value from their own data. Any genuine data ownership regime would collapse the current business model of Google, Meta, and most data brokers. **What would actually change things** - A statutory property right in personal data (proposed in various US bills, never passed) - Data intermediaries or cooperatives that negotiate collectively on your behalf - Personal data stores (Solid, MyData) -- technically viable, commercially marginal - MiCA-style sectoral regulation extending beyond financial data The honest answer: it has not changed because the entities that profit from the status quo have lobbying power far exceeding that of individual users, and the problem is diffuse enough that collective action is hard.

Tech

Dlaczego nie mogę otrzymać potwierdzenia, że moje dane zostały faktycznie usunięte?

Tech

Dlaczego nie mogę wiedzieć, czy to, co jest uruchomione, odpowiada temu, co deklarował mój SBOM?

Tech

Dlaczego każdy łańcuch proweniencji C2PA ulega zerwaniu w momencie, gdy treść trafia do mediów społecznościowych?

Tech

Dlaczego mogę przenosić swoje wpisy, ale nie profile, które platformy zbudowały na podstawie obserwowania mnie?