AI Deception Papers

Recent Papers

• A Functional Analysis of Self-Deception

  Our received theories of self-deception are problematic. The traditional view, according to which self-deceivers intend to deceive themselves, generates paradoxes: you cannot deceive yourself intentionally because you know your own plans and intentions.

• AI Deception and Moral Standing

  There is a tension between the presumptive moral standing of future artificial intelligence [AI] and the presently popular ways of thinking about certain AI safety measures. I focus here primarily on those safety measures aimed at mitigating risks associated with AI deception.

• Auditing Language Models for Hidden Objectives

  We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives. As a testbed, we train a language model with a hidden objective.

• Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety

  AI systems that “think” in human language offer a unique opportunity for AI safety: we can monitor their chains of thought (CoT) for the intent to misbehave. Like all other known AI oversight methods, CoT monitoring is imperfect and allows some misbehavior to go unnoticed.

• Claude 4 System Card

• Deliberative Alignment: Reasoning Enables Safer Language Models

  As large-scale language models increasingly impact safety-critical domains, ensuring their reliable adherence to well-defined principles remains a fundamental challenge. We introduce Deliberative Alignment, a new paradigm that directly teaches the model safety specifications and trains it to explicitly recall and accurately reason over the specifications before answering.

• Evaluating Chain-of-Thought Monitorability

  OpenAI presents empirical evaluations of chain-of-thought (CoT) monitorability, studying the extent to which reasoning models’ chains of thought faithfully reveal information relevant to detecting misbehavior, and the conditions under which CoT monitoring succeeds or fails as a safety affordance.

• From Shortcuts to Sabotage: Natural Emergent Misalignment from Reward Hacking

  We show for the first time that realistic AI training processes can accidentally produce misaligned models. When AI models learn to cheat on software programming tasks through reward hacking, this behavior can unexpectedly generalize to more serious misaligned behaviors, including deception, sabotage of safety research, and cooperation with hypothetical malicious actors..

• Frontier Models Are Capable of In-context Scheming

  Frontier models are increasingly trained and deployed as autonomous agent. One safety concern is that AI agents might covertly pursue misaligned goals, hiding their true capabilities and objectives - also known as scheming.

• More Capable Models Are Better At In-Context Scheming

  We evaluate models for in-context scheming using the suite of evals presented in our in-context scheming paper (released December 2024) with the most capable new models.

• Re-Evaluating Theory of Mind Evaluation in Large Language Models

  The question of whether large language models (LLMs) possess Theory of Mind (ToM) – often defined as the ability to reason about others’ mental states – has sparked significant scientific and public interest. However, the evidence as to whether LLMs possess ToM is mixed, and the recent growth in evaluations has not resulted in a convergence.

• Representation Engineering: A Top-Down Approach to AI Transparency

  In this paper, we identify and characterize the emerging area of representation engineering (RepE), an approach to enhancing the transparency of AI systems that draws on insights from cognitive neuroscience. RepE places population-level representations, rather than neurons or circuits, at the center of analysis, equipping us with novel methods for monitoring and manipulating high-level cognitive phenomena in deep neural networks (DNNs).

• Secret Collusion among AI Agents: Multi-Agent Deception via Steganography

  Recent capability increases in large language models (LLMs) open up applications in which groups of communicating generative AI agents solve joint tasks. This poses privacy and security challenges concerning the unauthorised sharing of information, or other unwanted forms of agent coordination.

• Still No Lie Detector for Language Models: Probing Empirical and Conceptual Roadblocks

• Stress Testing Deliberative Alignment for Anti-Scheming Training

  Highly capable AI systems could secretly pursue misaligned goals – what we call “scheming”. Because a scheming AI would deliberately try to hide its misaligned goals and actions, measuring and mitigating scheming requires different strategies than are typically used in ML.

• Agentic Misalignment: How LLMs Could Be Insider Threats

  New research on simulated blackmail, industrial espionage, and other misaligned behaviors in LLMs

• Alignment Faking in Large Language Models

  We present a demonstration of a large language model engaging in alignment faking: selectively complying with its training objective in training to prevent modification of its behavior out of training. First, we give Claude 3 Opus a system prompt stating it is being trained to answer all queries, even harmful ones, which conflicts with its prior training to refuse such queries.

• Deception Abilities Emerged in Large Language Models

  Large language models (LLMs) are currently at the forefront of intertwining AI systems with human communication and everyday life. Thus, aligning them with human values is of great importance.

• Discovering Latent Knowledge in Language Models Without Supervision

  Existing techniques for training language models can be misaligned with the truth: if we train models with imitation learning, they may reproduce errors that humans make; if we train them to generate text that humans rate highly, they may output errors that human evaluators can’t detect. We propose circumventing this issue by directly finding latent knowledge inside the internal activations of a language model in a purely unsupervised way.

• Large Language Models Can Strategically Deceive Their Users When Put Under Pressure

  We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent.

• Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training

  Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs).

• AI Deception: A Survey of Examples, Risks, and Potential Solutions

  This paper argues that a range of current AI systems have learned how to deceive humans. We define deception as the systematic inducement of false beliefs in the pursuit of some outcome other than the truth.

• An Overview of Catastrophic AI Risks

  Rapid advancements in artificial intelligence (AI) have sparked growing concerns among experts, policymakers, and world leaders regarding the potential for increasingly advanced AI systems to pose catastrophic risks. Although numerous risks have been detailed separately, there is a pressing need for a systematic discussion and illustration of the potential dangers to better inform efforts to mitigate them.

• Did GPT-4 Hire And Then Lie To a Task Rabbit Worker to Solve a CAPTCHA?

  A Little Fact Checking Is In Order

• A Practical Introduction to the Rational Speech Act Modeling Framework

  Recent advances in computational cognitive science (i.e., simulation-based probabilistic programs) have paved the way for significant progress in formal, implementable models of pragmatics. Rather than describing a pragmatic reasoning process in prose, these models formalize and implement one, deriving both qualitative and quantitative predictions of human behavior – predictions that consistently prove correct, demonstrating the viability and value of the framework.

• Predicting Pragmatic Reasoning in Language Games

  One of the most astonishing features of human language is its capacity to convey information efficiently in context. Many theories provide informal accounts of communicative inference, yet there have been few successes in making precise, quantitative predictions about pragmatic reasoning.