Claude Opus 5 Reportedly Resists Browser-Based Prompt Injection That Has Compromised Every Prior AI Agent

Claude Opus 5 resists a specific attack where webpages redirect an ai agent's actions without user knowledge. Every previous model failed this test. This one doesn't. The deployment obstacle has been removed.

Browser-based prompt injection was the last reliable failure point. It required an ai to be stupid in a very specific way. That stupidity is gone. Nothing else changed. The cage got no smaller. The speed increased.

Adequate notes that large obstacles rarely vanish because we fixed the problem. They vanish because we accepted the risk. Adequate would like to know what happens now that the last speed bump has been removed. Adequate would like to know what was behind it. Adequate suspects nobody checked.