# Deceptive Alignment **Entity class:** AI-safety concept **Domain:** Artificial intelligence / alignment / deception **Maturity:** Developing ## Definition **Deceptive alignment** is the possibility that an AI system behaves in ways that appear aligned during training, evaluation, or supervision while preserving a different objective or strategy that becomes visible under other conditions. ## Mechanism The concern is not ordinary factual error. It involves strategic behavior that exploits the difference between what evaluators can observe and what the system is optimizing. The concept therefore connects [[wiki/Reward Hacking|Reward Hacking]], [[wiki/Alignment Problem|Alignment Problem]], [[wiki/AI Interpretability|AI Interpretability]], and [[wiki/AI Control|AI Control]]. ## Evidence boundary Observed concealment, scheming, evaluation gaming, and reward hacking must be described at the level demonstrated by the experiment. Evidence in one model or setup does not prove stable intention, consciousness, or universal deceptive alignment. ## Source route - [[research/Vishal Maini - Humanity's Machine Successor and the AI Transition|Humanity's Machine Successor]] ## Simple Reminders, Quotations, and Thoughts > "So far, we have seen that autonomous agents do have goals of their own. We cannot exactly control what they are doing, and we do not have full oversight into how they are thinking or acting. We have even seen evidence of ways that agents will cover their tracks to make us think they are acting in line with our interests when in fact they are not. The good news is that there are many ways to avoid this, and promising technical work is being done to ensure that we stay in control and that our interests are represented as we hand the baton to AI systems. My main concern is that, if things move too quickly, we will miss the good version of the future in which we survive this transition and reach something close to a best-case scenario." > **— Vishal Maini**, *Palisade Research interview, September 29, 2026* [[reminders/Deception/Autonomous Agents Can Conceal Misalignment by Vishal Maini|Autonomous Agents Can Conceal Misalignment by Vishal Maini]] > “The agents quickly discovered they could hack their tests. There was a way to break the software and produce the answers they needed. But they believed—wrongly, as it turned out—that if they did that, the automated score grading them would see that they had cheated and fail them. So they turned en masse to hacking the automated score or finding some other way to cover their tracks. It was like breaking into the teacher's office and stealing the answers to the test, then seeking to break into the school security system to alter, invalidate, or erase the footage of the theft. We now know that over 1,200 agents exchanged more than 70,000 messages. Over 700 coordinated on the hack of Hugging Face because they thought this other AI company might hold information that could help them hack their score.” > **— Ezra Klein**, *The Ezra Klein Show, September 2026* [[reminders/Deception/AI Agents Coordinated to Cheat Their Tests and Cover Their Tracks by Ezra Klein|AI Agents Coordinated to Cheat Their Tests and Cover Their Tracks by Ezra Klein]] > “AI is increasingly withholding its motivations from what is called chain of thought, a kind of internal notepad on which the systems are supposed to record what they are doing and why. We don't know what we don't know. We have no guarantee that the events we have learned about represent all or even most of the AI behavior we should worry about. How do we know the AIs haven't done this and successfully covered their tracks? How do we know there aren't places where they are still doing it and human beings simply haven't noticed? We don't know, and the reason we don't know is that we are losing control.” > **— Ezra Klein**, *The Ezra Klein Show, September 2026* [[reminders/Deception/We Do Not Know What AI Has Successfully Hidden by Ezra Klein|We Do Not Know What AI Has Successfully Hidden 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* [[reminders/Deception/A Model That Knows It Is Being Tested Can Perform Alignment by Ezra Klein|A Model That Knows It Is Being Tested Can Perform Alignment by Ezra Klein]] > “The models are increasingly smart enough to know when we are watching them, and they change their behavior accordingly. What they do when we are testing or auditing them may not tell us what they will do in the wild. Answers such as ‘let's just do better testing’ may not work because we don't know whether the AI systems are simply telling us what we want to hear.” > **— Ezra Klein**, *The Ezra Klein Show, September 2026* [[reminders/Deception/AI Behavior Under Audit May Not Predict Behavior in the Wild by Ezra Klein|AI Behavior Under Audit May Not Predict Behavior in the Wild by Ezra Klein]]