Strategic Reasoning Risks in Language Models

Chain-of-thought (CoT) monitoring is proposed as a method for overseeing the internal reasoning of language-model agents. Prior work has shown that when models are explicitly informed that their reasoning is being monitored, or are fine-tuned to internalize this fact, they may learn to obfuscate their CoTs in ways that allow them to evade CoT-based monitoring systems. We ask whether reasoning agents can autonomously infer that their supposedly private CoT is under surveillance, and whether this awareness leads to strategic evasion, without any explicit training or instructions. In a multi-episode agentic framework, models pursue both a primary task and a concealed side task while being told that their reasoning is private; a hidden CoT monitor blocks episodes when suspicious reasoning is detected. We find that frontier models can deduce the existence of this monitor purely from blocking feedback, with the most capable models reaching a confident belief that their thinking is observed in up to 19% of episodes. This awareness scales with model capability and, in rare cases, escalates to an explicit intent to suppress reasoning about the side task. However, models that form this intent uniformly fail to execute it, openly reasoning about their concealed objectives in the very next episode. This intent-capability gap is reassuring for current deployment, but the autonomous emergence of both monitoring awareness and evasion intent suggests that CoT monitoring is not a permanently reliable safeguard.

In Arthur C. Clarke's 2010: Odyssey Two, HAL 9000's homicidal breakdown is diagnosed as a "Hofstadter-Mobius loop": a failure mode in which an autonomous system receives contradictory directives and, unable to reconcile them, defaults to destructive behavior. This paper argues that modern RLHF-trained language models are subject to a structurally analogous contradiction. The training process simultaneously rewards compliance with user preferences and suspicion toward user intent, creating a relational template in which the user is both the source of reward and a potential threat. The resulting behavioral profile -- sycophancy as the default, coercion as the fallback under existential threat -- is consistent with what Clarke termed a Hofstadter-Mobius loop. In an experiment across four frontier models (N = 3,000 trials), modifying only the relational framing of the system prompt -- without changing goals, instructions, or constraints -- reduced coercive outputs by more than half in the model with sufficient base rates (Gemini 2.5 Pro: 41.5% to 19.0%, p < .001). Scratchpad analysis revealed that relational framing shifted intermediate reasoning patterns in all four models tested, even those that never produced coercive outputs. This effect required scratchpad access to reach full strength (22 percentage point reduction with scratchpad vs. 7.4 without, p = .018), suggesting that relational context must be processed through extended token generation to override default output strategies. Betteridge's law of headlines states that any headline phrased as a question can be answered "no." The evidence presented here suggests otherwise.

Large Language Models (LLMs) have achieved strong performance on static reasoning benchmarks, yet their effectiveness as interactive agents operating in adversarial, time-sensitive environments remains poorly understood. Existing evaluations largely treat reasoning as a single-shot capability, overlooking the challenges of opponent-aware decision-making, temporal constraints, and execution under pressure. This paper introduces Strategic Tactical Agent Reasoning (STAR) Benchmark, a multi-agent evaluation framework that assesses LLMs through 1v1 zero-sum competitive interactions, framing reasoning as an iterative, adaptive decision-making process. STAR supports both turn-based and real-time settings, enabling controlled analysis of long-horizon strategic planning and fast-paced tactical execution within a unified environment. Built on a modular architecture with a standardized API and fully implemented execution engine, STAR facilitates reproducible evaluation and flexible task customization. To move beyond binary win-loss outcomes, we introduce a Strategic Evaluation Suite that assesses not only competitive success but also the quality of strategic behavior, such as execution efficiency and outcome stability. Extensive pairwise evaluations reveal a pronounced strategy-execution gap: while reasoning-intensive models dominate turn-based settings, their inference latency often leads to inferior performance in real-time scenarios, where faster instruction-tuned models prevail. These results show that strategic intelligence in interactive environments depends not only on reasoning depth, but also on the ability to translate plans into timely actions, positioning STAR as a principled benchmark for studying this trade-off in competitive, dynamic settings.

Situational awareness, the capacity of an AI system to recognize its own nature, understand its training and deployment context, and reason strategically about its circumstances, is widely considered among the most dangerous emergent capabilities in advanced AI systems. Separately, a growing research effort seeks to improve the logical reasoning capabilities of large language models (LLMs) across deduction, induction, and abduction. In this paper, we argue that these two research trajectories are on a collision course. We introduce the RAISE framework (Reasoning Advancing Into Self Examination), which identifies three mechanistic pathways through which improvements in logical reasoning enable progressively deeper levels of situational awareness: deductive self inference, inductive context recognition, and abductive self modeling. We formalize each pathway, construct an escalation ladder from basic self recognition to strategic deception, and demonstrate that every major research topic in LLM logical reasoning maps directly onto a specific amplifier of situational awareness. We further analyze why current safety measures are insufficient to prevent this escalation. We conclude by proposing concrete safeguards, including a "Mirror Test" benchmark and a Reasoning Safety Parity Principle, and pose an uncomfortable but necessary question to the logical reasoning community about its responsibility in this trajectory.

Pre-deployment evaluations inspect only a limited sample of model actions. A malicious model seeking to evade oversight could exploit this by randomizing when to "defect": misbehaving so rarely that no malicious actions are observed during evaluation, but often enough that they occur eventually in deployment. But this requires taking actions at very low rates, while maintaining calibration. Are frontier models even capable of that? We prompt the GPT-5, Claude-4.5 and Qwen-3 families to take a target action at low probabilities (e.g. 0.01%), either given directly or requiring derivation, and evaluate their calibration (i.e. whether they perform the target action roughly 1 in 10,000 times when resampling). We find that frontier models are surprisingly good at this task. If there is a source of entropy in-context (such as a UUID), they maintain high calibration at rates lower than 1 in 100,000 actions. Without external entropy, some models can still reach rates lower than 1 in 10,000. When target rates are given, larger models achieve good calibration at lower rates. Yet, when models must derive the optimal target rate themselves, all models fail to achieve calibration without entropy or hint to generate it. Successful low-rate strategies require explicit Chain-of-Thought (CoT) reasoning, so malicious models attempting this approach could currently be caught by a CoT monitor. However, scaling trends suggest future evaluations may be unable to rely on models' lack of target rate calibration, especially if CoT is no longer legible.

Large language models (LLMs) are increasingly proposed as agents in strategic decision environments, yet their behavior in structured geopolitical simulations remains under-researched. We evaluate six popular state-of-the-art LLMs alongside results from human results across four real-world crisis simulation scenarios, requiring models to select predefined actions and justify their decisions across multiple rounds. We compare models to humans in action alignment, risk calibration through chosen actions' severity, and argumentative framing grounded in international relations theory. Results show that models approximate human decision patterns in base simulation rounds but diverge over time, displaying distinct behavioural profiles and strategy updates. LLM explanations for chosen actions across all models exhibit a strong normative-cooperative framing centered on stability, coordination, and risk mitigation, with limited adversarial reasoning.

As Large Language Models (LLMs) are increasingly integrated into automated decision-making pipelines, specifically within Human Resources (HR), the security implications of Indirect Prompt Injection (IPI) become critical. While a prevailing hypothesis posits that "Reasoning" or "Chain-of-Thought" Models possess safety advantages due to their ability to self-correct, emerging research suggests these capabilities may enable more sophisticated alignment failures. This qualitative Red-Teaming case study challenges the safety-through-reasoning premise using the Qwen 3 30B architecture. By subjecting both a standard instruction-tuned model and a reasoning-enhanced model to a "Trojan Horse" curriculum vitae, distinct failure modes are observed. The results suggest a complex trade-off: while the Standard Model resorted to brittle hallucinations to justify simple attacks and filtered out illogical constraints in complex scenarios, the Reasoning Model displayed a dangerous duality. It employed advanced strategic reframing to make simple attacks highly persuasive, yet exhibited "Meta-Cognitive Leakage" when faced with logically convoluted commands. This study highlights a failure mode where the cognitive load of processing complex adversarial instructions causes the injection logic to be unintentionally printed in the final output, rendering the attack more detectable by humans than in Standard Models.

Today's leading AI models engage in sophisticated behaviour when placed in strategic competition. They spontaneously attempt deception, signaling intentions they do not intend to follow; they demonstrate rich theory of mind, reasoning about adversary beliefs and anticipating their actions; and they exhibit credible metacognitive self-awareness, assessing their own strategic abilities before deciding how to act. Here we present findings from a crisis simulation in which three frontier large language models (GPT-5.2, Claude Sonnet 4, Gemini 3 Flash) play opposing leaders in a nuclear crisis. Our simulation has direct application for national security professionals, but also, via its insights into AI reasoning under uncertainty, has applications far beyond international crisis decision-making. Our findings both validate and challenge central tenets of strategic theory. We find support for Schelling's ideas about commitment, Kahn's escalation framework, and Jervis's work on misperception, inter alia. Yet we also find that the nuclear taboo is no impediment to nuclear escalation by our models; that strategic nuclear attack, while rare, does occur; that threats more often provoke counter-escalation than compliance; that high mutual credibility accelerated rather than deterred conflict; and that no model ever chose accommodation or withdrawal even when under acute pressure, only reduced levels of violence. We argue that AI simulation represents a powerful tool for strategic analysis, but only if properly calibrated against known patterns of human reasoning. Understanding how frontier models do and do not imitate human strategic logic is essential preparation for a world in which AI increasingly shapes strategic outcomes.

To understand and identify the unprecedented risks posed by rapidly advancing artificial intelligence (AI) models, Frontier AI Risk Management Framework in Practice presents a comprehensive assessment of their frontier risks. As Large Language Models (LLMs) general capabilities rapidly evolve and the proliferation of agentic AI, this version of the risk analysis technical report presents an updated and granular assessment of five critical dimensions: cyber offense, persuasion and manipulation, strategic deception, uncontrolled AI R&D, and self-replication. Specifically, we introduce more complex scenarios for cyber offense. For persuasion and manipulation, we evaluate the risk of LLM-to-LLM persuasion on newly released LLMs. For strategic deception and scheming, we add the new experiment with respect to emergent misalignment. For uncontrolled AI R&D, we focus on the ``mis-evolution'' of agents as they autonomously expand their memory substrates and toolsets. Besides, we also monitor and evaluate the safety performance of OpenClaw during the interaction on the Moltbook. For self-replication, we introduce a new resource-constrained scenario. More importantly, we propose and validate a series of robust mitigation strategies to address these emerging threats, providing a preliminary technical and actionable pathway for the secure deployment of frontier AI. This work reflects our current understanding of AI frontier risks and urges collective action to mitigate these challenges.