other
Security research venue credited with a study of sixteen code-generation models across more than 500,000 samples; the study itself is not named or linked in the source.
No evidence-backed relationships are recorded.
USENIX Security 2025 researchers found 19.7% of packages suggested by 16 LLMs were fake, and 43% of those names recurred on every re-run. Names that repeat can be registered ahead of time, so a team has to vet a suggested dependency before installing it, even when the install succeeds.
Reality
UMass Amherst researchers bought groceries with an expired contactless card. The date that is supposed to kill it is unsigned, zeroed out before the issuer sees it, and checked by nobody in particular.
Perspective Coverage
5 publishersReality
CSAIL researchers timed an interrupt to land inside the gap AMD's saferet defense leaves open, then pulled the root password file off a current Linux kernel. AMD has shipped a mitigation. Intel's picture is messier.
Reality
Academics rebuilt Tiangou Secure Gateway firmware from leaked Git history and matched its filtering to the Great Firewall. One of three DNS injectors matched, and that ratio is the interesting part.
Reality
Academics rebuilt Geedge Networks' Tiangou Secure Gateway firmware from a 100,000-file leak and matched its filtering to China's censorship system. The same box has been exported.
Reality
A vendor essay on AI package hallucination makes a defensible case: a package name that does not exist yet cannot be scanned, so the control has to sit at selection.
Reality