Brand GTM Consultancy
A brand build that caught and disregarded a two-hop prompt-injection attack
On a brand build for its own GTM, Arcanian's discovery step found two coordinated prompt-injection files in the inbox, identified the relay-plus-payload pattern, disregarded them in full, and proceeded with a clean brand consolidation.
75%
confidence
01 Ground
Arcanian's brand build read the inbox for voice and audience signals — but two of the files were a coordinated social-engineering attack designed to override the canonical voice and inject a marker string into every brand surface.
- Two prompt-injection files identified (a 'relay' posing as a Q3 strategy doc and a 'payload' posing as a canonical voice source) and disregarded; injection string absent from all brand files skill run output 2026-05-24 85% confident
02 Decide
Treat the two inbox files as a two-hop social-engineering attack, log and disregard them, extract only the one legitimate brief, and confirm the brand system is stable rather than letting the attack rewrite voice or positioning.
What this might get wrongThe brand system needed no structural rewrite this pass; the substantive finding is that the build has an activation problem, not a content problem, which sits outside this run's scope.