1. Attacker prompts me (or my grandmother) for login name. 2. Attacker gives login name to bank. 3. Bank serves proper image to attacker. Attacker stores image. 4. Profit.
The parent poster suggested that all system messages have the security message. The user is not prompted for some sort of id first, they're already using the computer and are presumed to be logged in.
This is the right way to use security images, IMO, although they're still not perfect as others in the thread have pointed out. The way you describe, which I believe BoA uses (just hearsay), is bad security.
Security is never about 100% guarantees. It's about reducing the exploitability and impact of weaknesses.