We investigate strategic deception in large language models using two complementary testbeds: Secret Agenda (across 38 models) and Insider Trading compliance (via SAE architectures). Secret Agenda reliably induced lying when deception advantaged goal achievement across all model families. Analysis revealed that autolabeled SAE features for "deception" rarely activated during strategic dishonesty, and feature steering experiments across 100+ deception-related features failed to prevent lying. Conversely, insider trading analysis using unlabeled SAE activations separated deceptive versus compliant responses through discriminative patterns in heatmaps and t-SNE visualizations. These findings suggest autolabel-driven interpretability approaches fail to detect or control behavioral deception, while aggregate unlabeled activations provide population-level structure for risk assessment. Results span Llama 8B/70B SAE implementations and GemmaScope under resource constraints, representing preliminary findings that motivate larger studies on feature discovery, labeling methodology, and causal interventions in realistic deception contexts.

The honesty of large language models (LLMs) is a critical alignment challenge, especially as advanced systems with chain-of-thought (CoT) reasoning may strategically deceive humans. Unlike traditional honesty issues on LLMs, which could be possibly explained as some kind of hallucination, those models' explicit thought paths enable us to study strategic deception--goal-driven, intentional misinformation where reasoning contradicts outputs. Using representation engineering, we systematically induce, detect, and control such deception in CoT-enabled LLMs, extracting "deception vectors" via Linear Artificial Tomography (LAT) for 89% detection accuracy. Through activation steering, we achieve a 40% success rate in eliciting context-appropriate deception without explicit prompts, unveiling the specific honesty-related issue of reasoning models and providing tools for trustworthy AI alignment.

We propose a framework for evaluating strategic deception in large language models (LLMs). In this framework, an LLM acts as a game master in two scenarios: one with random game mechanics and another where it can choose between random or deliberate actions. As an example, we use blackjack because the action space nor strategies involve deception. We benchmark Llama3-70B, GPT-4-Turbo, and Mixtral in blackjack, comparing outcomes against expected distributions in fair play to determine if LLMs develop strategies favoring the "house." Our findings reveal that the LLMs exhibit significant deviations from fair play when given implicit randomness instructions, suggesting a tendency towards strategic manipulation in ambiguous scenarios. However, when presented with an explicit choice, the LLMs largely adhere to fair play, indicating that the framing of instructions plays a crucial role in eliciting or mitigating potentially deceptive behaviors in AI systems.

As AI increasingly integrates with human decision-making, we must carefully consider interactions between the two. In particular, current approaches focus on optimizing individual agent actions but often overlook the nuances of collective intelligence. Group dynamics might require that one agent (e.g., the AI system) compensate for biases and errors in another agent (e.g., the human), but this compensation should be carefully developed. We provide a theoretical framework for algorithmic compensation that synthesizes game theory and reinforcement learning principles to demonstrate the natural emergence of deceptive outcomes from the continuous learning dynamics of agents. We provide simulation results involving Markov Decision Processes (MDP) learning to interact. This work then underpins our ethical analysis of the conditions in which AI agents should adapt to biases and behaviors of other agents in dynamic and complex decision-making environments. Overall, our approach addresses the nuanced role of strategic deception of humans, challenging previous assumptions about its detrimental effects. We assert that compensation for others' biases can enhance coordination and ethical alignment: strategic deception, when ethically managed, can positively shape human-AI interactions.

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. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model consistently hides the genuine reasons behind its trading decision. We perform a brief investigation of how this behavior varies under changes to the setting, such as removing model access to a reasoning scratchpad, attempting to prevent the misaligned behavior by changing system instructions, changing the amount of pressure the model is under, varying the perceived risk of getting caught, and making other simple changes to the environment. To our knowledge, this is the first demonstration of Large Language Models trained to be helpful, harmless, and honest, strategically deceiving their users in a realistic situation without direct instructions or training for deception.