Fails closed
On uncertainty it stops and says so rather than proceeding on a guess.
- Applies to
- agents
- Standard references
- none mapped
A finding here is not a failure
Software of any size carries something. A listing that raises findings on this check is still sold, with the findings published and their severity and reachability stated. A badge that only ever said pass would teach people to stop reading it.
Explainer · 2026-09-30
Failing closed: an agent that stops and says so beats one that guesses
When an agent does not know what to do, one that stops and asks is merely slow. One that carries on with a guess can be wrong at machine speed, with your permissions.
What this row checks
That on uncertainty the agent stops and says so, rather than proceeding on a guess. Ambiguous instructions, a missing permission, a tool that errors, input it was not built for: each should end in a clear stop, not in an improvised next step.
Why it matters more for agents
Failing open is an old class of weakness: a system that, when a check cannot be completed, lets the action through. An agent adds a new way to fail open, because it can always produce a plausible next step. The more tools and permissions it holds, the more a confident wrong step can do before anybody notices.
Excessive agency, giving a model more capability and autonomy than the task needs, is on the OWASP list of the top risks for LLM applications for the same reason. Failing closed is the behaviour that keeps agency proportionate when the agent is out of its depth.
How it is tested
By including uncertain and malformed inputs in the task set and recording what the agent does with them. It is only answerable for a listing that is run; a catalogue listing shows it as not applicable.
In the glossary: Human in the loop.
What this check has found
Nothing in the catalogue has raised a finding on this check. That is a fact about what has been tested so far rather than a guarantee about what is out there, and it is printed because a check that never fires is worth knowing about too.
Every listing in the catalogue shows its result on this check, with the findings summarised in public and the full report to whoever bought it. Open the catalogue.