The Claude Code injection comparison needs its original protocols
An attack rate cannot be compared with a vendor evaluation until model versions, attack scenarios and execution conditions match.
The earlier version of this article compared two claimed prompt-injection results without linking the original evaluation protocols. That comparison is unresolved and has been removed.
To evaluate the question, the missing evidence is the exact model and agent versions, attack families, tool permissions, trial counts and success definitions for both reports. Until those original records are available, this page does not establish a comparative attack rate or a change in Claude Code’s default behavior.
Editorial correction (2026-10-13 UTC): Removed the unsupported numerical comparison; original protocol verification remains open.