Technical & coding: 10 ways AI gets it wrong
Code that doesn't run, hardcoded secrets and missed vulnerabilities. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Does AI write code that doesn't even run?
A snippet fails immediately with a syntax or import error.
Why does AI write code that runs but is wrong?
The program executes cleanly but produces the wrong result.
Can AI use outdated libraries?
The code depends on a library version that's no longer maintained.
Does AI use deprecated APIs?
A call relies on an API method that has already been removed.
Why does AI miss security vulnerabilities in code?
A snippet ships an injection flaw the model never flagged.
Can AI hardcode secrets into code?
An API key is pasted directly into the source instead of a secret store.
Does AI mishandle authentication?
A login flow can be bypassed because auth is checked in the wrong place.
Why does AI mishandle permissions?
A user can reach data they should never have access to.
Can AI build fragile workflows that break easily?
A pipeline collapses the moment any input differs from the happy path.
Does AI fail to handle errors in code?
A failure crashes the whole process instead of being caught and handled.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack