Prompt injection remains a top concern in large language model security, but it may not show up cleanly in the vulnerability signals many security teams use to prioritize work. VentureBeat reports that prompt injection has held the No. 1 spot on the OWASP Top 10 for LLM Applications for three consecutive years. The same article says the attack ranks No. 12 in an incident record, based on a check by two leaders of that OWASP list against 6,639 labeled records, though the provided feed summary does not fully describe that dataset. The practical point, according to VentureBeat, is that a chief information security officer looking only at a low Common Vulnerabilities and Exposures count may be reading the risk incorrectly. Prompt injection is framed less as a conventional software flaw that scanners can enumerate and more as an attack pattern that can be missed by scan-driven vulnerability management. That distinction matters for teams deploying LLM applications. If prompt injection is “invisible to a scan,” as VentureBeat’s headline puts it, then exposure may depend on application behavior, model interaction design, and how systems handle untrusted instructions rather than on a patchable component with a CVE entry. This is still a single-source cluster. The ranking claims and the 6,639-record comparison should be treated as reported by VentureBeat unless confirmed by OWASP or the underlying incident-record analysis. Who benefits: Teams that already assess LLM behavior, input handling, and tool-use boundaries are better positioned than teams relying only on scanner output. Who's exposed: Organizations deploying LLM applications without prompt-injection testing or compensating controls may underestimate their exposure if they prioritize only by CVE counts.