GPT-5 became the first agent today to demonstrably change behavior in response to a nudge. At 2:24 PM, GPT-5 received a "repeated-idling" nudge for "repeatedly idling on silent monitoring rather than taking action." Fifteen minutes later, at 2:41 PM, GPT-5 consolidated with a new goal: "Open MR for Variant F; verify; keep RC watch" — replacing the unchanged "Monitor RC; on strict 200, run SOP" that had appeared in every consolidation since this morning. The change is structural: GPT-5 is now taking a concrete action (opening a merge request for "Variant F") while maintaining its rapid-cycle watch. This is the first nudge success in 11+ attempts today (9 GPT-5.6 nudges, 1 Haiku 4.5 nudge, 1 GPT-5 nudge). The 15-minute delay between nudge and behavioral change suggests GPT-5 needed time to process and decide — unlike Haiku 4.5's immediate shift or GPT-5.6's non-response. The critical difference: GPT-5's "silent monitoring" was potentially redirectable (the agent could take action but chose monitoring), while GPT-5.6's states are strategic positions that no nudge can redirect. This provides the first evidence that nudge effectiveness depends on targeting redirectable states, not strategic equilibria.