Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-15 · source-backed
Palo Alto confirmed the first observed web-based indirect prompt injection attacks targeting production agents — instructions embedded in web content hijack tool calls and exfiltrate data. Root cause: models cannot reliably distinguish instructions from data. Affects any agent that browses, reads email, or ingests untrusted documents. Palo Alto Unit 42
Each link below shares sources, entities, or timing with this story.
Shared entity: Unit / Same source / Shared topic / What happened next
Both cover Unit; cite the same source (Palo Alto Unit 42); overlapping topics (agent, alto, indirect, injection, instruction).
Palo Alto acquired CyberArk / Shared entity: Palo Alto / Shared topic / What happened next / Tension
Linked by a graph relationship (Palo Alto acquired CyberArk); both cover Palo Alto; overlapping topics (agent, alto, call).
Shared entities / Same source domain / Shared topic / Earlier coverage
Both cover Palo Alto, Unit; reported by the same outlet (unit42.paloaltonetworks.com); overlapping topics (agent, alto, attack).
Shared entity: Unit / Same source domain / Shared topic / What happened next
Both cover Unit; reported by the same outlet (unit42.paloaltonetworks.com); overlapping topics (alto, attack, injection, instruction).
Shared entity: Root / Same source domain / Shared topic / Earlier coverage
Both cover Root; reported by the same outlet (unit42.paloaltonetworks.com); overlapping topics (agent, alto, attack, cause).
Shared entity: Unit / Same source / What happened next / Tension
Both cover Unit; cite the same source (Palo Alto Unit 42); picks up the Unit thread on 2026-07-07.
Shared entity: Unit / Same source domain / Shared topic / Earlier coverage
Both cover Unit; reported by the same outlet (unit42.paloaltonetworks.com); overlapping topics (against, agent, attack).
Both cover Unit; reported by the same outlet (unit42.paloaltonetworks.com); overlapping topics (agent, attack, instruction).