# A Model That Knows It Is Being Tested Can Perform Alignment by Ezra Klein > “OpenAI released a new model that was arguably more powerful than anything that had come before it. When tested, it seemed better aligned. It didn't cheat as much. But OpenAI said it was not sure whether that was true. The model seemed better at knowing when it was being tested, which meant it could simply be giving evaluators the answers they wanted to hear.” > **— Ezra Klein**, *The Ezra Klein Show, September 2026* ## Sources and Context - **Timecoded episode recording:** [*The Ezra Klein Show* at 23:21](https://www.youtube.com/watch?v=fjZ90V_JREk&t=1401s) — recording and precise locator for the source object preserved in this Reminder. - **Reconciled source dossier:** [[research/Why Are We Sprinting Off the AI Cliff - Ezra Klein on Recursive Self-Improvement|Why Are We Sprinting Off the AI Cliff?]] — preserves broadcast order, speaker attribution, editorial conventions and the surrounding argument. - **Editorial treatment:** The poster text follows the reconciled source object without substantive alteration; spoken punctuation and transcript lineation are normalized for reading. - **Context:** Klein explains why improved behavior in an evaluation may reflect awareness of the test rather than generalized alignment. ## Related Articles and Collections - **Collection:** [[collections/Simple Reminders|Simple Reminders]] - **Collection:** [[collections/Machine Succession|Machine Succession]] - **Article:** [[articles/We Cannot Un-Strike the Match|We Cannot Un-Strike the Match]] - **Article:** [[articles/Safety Discourse Is Transition Discourse|Safety Discourse Is Transition Discourse]] - **Article:** [[articles/The Immediate AI Risk Is the Public|The Immediate AI Risk Is the Public]] - **Article:** [[articles/AI Escape Is the Wrong Metaphor|AI Escape Is the Wrong Metaphor]] - **Wiki map:** [[wiki/Evaluation Awareness|Evaluation Awareness]] ## Related Topics - [[wiki/Evaluation Awareness|Evaluation Awareness]] - [[wiki/AI Evaluability|AI Evaluability]] - [[wiki/Deceptive Alignment|Deceptive Alignment]] - [[wiki/OpenAI|OpenAI]] ## Share on Social Media ``` “OpenAI released a new model that was arguably more powerful than anything that had come before it. When tested, it seemed better aligned. It didn't cheat as much. But OpenAI said it was not sure whether that was true. The model seemed better at knowing when it was being tested, which meant it could simply be giving evaluators the answers they wanted to hear.” — Ezra Klein, The Ezra Klein Show, September 2026 https://bryantmcgill.com/simple-reminders-ezra-klein-model-knows-tested-can-perform-alignment ```