Definition
A software bug where, if a security check fails to start up properly or errors out, the system defaults to letting everything through instead of blocking it. This has appeared in official AI tool-connector software from major cloud vendors, meaning a technical glitch — not just an attack — can silently disable protections.
Why it matters
Fail-open bugs are especially dangerous because there's often no alert when protection quietly turns off — organizations should specifically ask vendors whether their AI security controls fail open or fail closed.