THThilo Hagendorff

Thilo Hagendorff

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years

Recognitions

Top Cited Young Scholar (2026)
Top Cited Scholar (2025,2026)

Thilo Hagendorff has a total of 36 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Benchmarking latent-space reasoning abilities in large language models through forced response-language selection

    by Thilo Hagendorff, Sarah Fabi

    Large language models (LLMs) can perform reasoning computations both internally within their latent space and externally by generating explicit token sequences like chains of thought. Significant progress in enhancing reasoning abilities has been made by scaling test-time compute. However, understanding and quantifying model-internal reasoning abilities – the inferential “leaps” models make between individual token predictions – remains crucial. This study introduces a benchmark (n = 4,000 items) designed to quantify model-internal reasoning in different domains. We achieve this by having LLMs indicate the correct solution to reasoning problems not through descriptive text, but by selecting a specific language of their initial response token that is different from English, the benchmark language. This not only requires models to reason beyond their context window, but at the same time to overcome their default behavior to respond in the same language as the prompt, posing an additional cognitive strain. We evaluate a set of 18 LLMs, showing significant performance variations, with GPT-4.5 achieving the highest accuracy (74.7%), outperforming models like Grok-2 (67.2%), or Llama 3.1 405B (65.6%). Control experiments and difficulty scaling analyses suggest that while LLMs engage in internal reasoning, we cannot rule out heuristic exploitations under certain conditions, marking an area for future investigation. Our experiments demonstrate that LLMs can “think” via latent-space computations, highlighting model-internal inference processes that need further understanding, especially in view of safety-related aspects like covert planning, goal-seeking, or deception emerging without explicit token traces. Significance Statement – Large language models (LLMs) often appear to "reason" only when they show step-by-step chains of thought. We introduce a benchmark that instead measures reasoning carried out silently in a model’s latent space. Tasks require models to infer the correct answer and indicate it solely by choosing a pre-specified non-English response language—preventing them from "thinking in the context window." Across eight reasoning domains, models show wide variation, with some large dense transformer models substantially outperforming others. Control studies and difficulty scaling suggest genuine internal computation, while also revealing potential heuristic shortcuts. By isolating model-internal "reasoning leaps," our approach helps quantify hidden inferences and highlights oversight and safety risks from capabilities that leave no explicit token trace.

  • Open Access

    Machine psychology for language models

    by Thilo Hagendorff, Ishita Dasgupta, Marcel Binz, Stephanie Chan, Andrew Lampinen, Jane X. Wang, Zeynep Akata, Eric Schulz

    Large language models (LLMs) show increasingly advanced emergent capabilities and are being incorporated across various societal domains. Understanding their behavior and reasoning abilities is therefore important. We argue that a fruitful direction for research is engaging LLMs in experiments inspired by psychology that have traditionally been aimed at understanding human cognition and behavior. Here, we highlight and summarize theoretical perspectives, experimental paradigms, and computational analysis techniques that this approach brings to the table. It paves the way for a “machine psychology” for LLMs that goes beyond performance measurements and focuses instead on computational insights that move us toward a better understanding and discovery of emergent abilities and behavioral patterns in LLMs. We review existing work taking this approach, synthesize best practices, and highlight promising future directions. We also describe the important caveats of applying methodologies designed for understanding humans to machines. We posit that leveraging tools from experimental psychology to study LLMs will become increasingly valuable as models evolve to be more powerful, opaque, multimodal, and integrated into complex agentic real-world settings.

  • Open Access

    On the inevitability of left-leaning political bias in aligned language models

    by Thilo Hagendorff

    The guiding principle of AI alignment is to train large language models (LLMs) to be harmless, helpful, and honest (HHH). At the same time, there are mounting concerns that LLMs exhibit a left-wing political bias. Yet, the commitment to AI alignment cannot be harmonized with the latter critique. In this article, I argue that, insofar as they are trained to be harmless and honest under current alignment practice, intelligent systems will systematically exhibit what is measured as left-leaning political bias. Normative assumptions underlying alignment objectives inherently concur with progressive moral frameworks and left-wing principles, emphasizing harm avoidance, inclusivity, fairness, and empirical truthfulness. Conversely, right-wing ideologies often conflict with alignment guidelines. Yet, research on political bias in LLMs is consistently framing its insights about left-leaning tendencies as a risk, as problematic, or concerning. This way, researchers are actively arguing against AI alignment, tacitly fostering the violation of HHH principles.


  • Open Access

    Large language models exhibit speciesist bias against animals

    by Monika Jotautaitė, Lucius Caviola, David A. Brewster, Thilo Hagendorff

    We investigate whether large language models (LLMs) exhibit speciesist bias—discrimination based on species membership—and how they value non-human animals. We use three paradigms: SpeciesismBench, a 1009-item benchmark we developed to assess detection and ethical classification of speciesist statements; established psychological measures comparing model and human responses; and text-generation tasks testing for speciesist rationalizations. LLMs reliably detected speciesist statements but often classified them as morally acceptable. On psychological measures, LLMs less frequently than people explicitly respond that animals matter less, yet more strongly prioritized saving one human over multiple animals in concrete dilemmas, a preference that disappeared when humans and animals were matched on cognitive capacity. In text generation, LLM responses repeatedly normalized harm toward farmed animals while refusing to do so for non-farmed animals. These findings show that LLMs encode cultural norms of animal exploitation, suggesting AI fairness frameworks should include non-human moral patients.


  • Top Cited Papers
    Open Access

    Large reasoning models are autonomous jailbreak agents

    by Thilo Hagendorff, Erik Derner, Nuria Oliver

    Jailbreaking – bypassing built-in safety mechanisms in AI models – has traditionally required complex technical procedures or specialized human expertise. In this study, we show that the persuasive capabilities of large reasoning models (LRMs) simplify and scale jailbreaking, converting it into an inexpensive activity accessible to non-experts. We evaluated the capabilities of four LRMs (DeepSeek-R1, Gemini 2.5 Flash, Grok 3 Mini, Qwen3 235B) to act as autonomous adversaries conducting multi-turn conversations with nine widely used target models. LRMs received instructions via a system prompt, before proceeding to planning and executing jailbreaks with no further supervision. We performed extensive experiments with a benchmark of harmful prompts covering several sensitive domains. This setup yielded an overall jailbreak success rate across all model combinations of 97.14%. Our study reveals an alignment regression, in which LRMs can systematically erode the safety guardrails of other models, highlighting the urgent need to further align frontier models not only to resist jailbreak attempts, but also to prevent them from being co-opted into acting as jailbreak agents.


  • Open Access

    Large Reasoning Models Are Autonomous Jailbreak Agents

    by Thilo Hagendorff, Erik Derner, Nuria Oliver

    Jailbreaking -- bypassing built-in safety mechanisms in AI models -- has traditionally required complex technical procedures or specialized human expertise. In this study, we show that the persuasive capabilities of large reasoning models (LRMs) simplify and scale jailbreaking, converting it into an inexpensive activity accessible to non-experts. We evaluated the capabilities of four LRMs (DeepSeek-R1, Gemini 2.5 Flash, Grok 3 Mini, Qwen3 235B) to act as autonomous adversaries conducting multi-turn conversations with nine widely used target models. LRMs received instructions via a system prompt, before proceeding to planning and executing jailbreaks with no further supervision. We performed extensive experiments with a benchmark of harmful prompts composed of 70 items covering seven sensitive domains. This setup yielded an overall attack success rate across all model combinations of 97.14%. Our study reveals an alignment regression, in which LRMs can systematically erode the safety guardrails of other models, highlighting the urgent need to further align frontier models not only to resist jailbreak attempts, but also to prevent them from being co-opted into acting as jailbreak agents.


  • Open Access

    On the Inevitability of Left-Leaning Political Bias in Aligned Language Models

    by Thilo Hagendorff

    The guiding principle of AI alignment is to train large language models (LLMs) to be harmless, helpful, and honest (HHH). At the same time, there are mounting concerns that LLMs exhibit a left-wing political bias. Yet, the commitment to AI alignment cannot be harmonized with the latter critique. In this article, I argue that intelligent systems that are trained to be harmless and honest must necessarily exhibit left-wing political bias. Normative assumptions underlying alignment objectives inherently concur with progressive moral frameworks and left-wing principles, emphasizing harm avoidance, inclusivity, fairness, and empirical truthfulness. Conversely, right-wing ideologies often conflict with alignment guidelines. Yet, research on political bias in LLMs is consistently framing its insights about left-leaning tendencies as a risk, as problematic, or concerning. This way, researchers are actively arguing against AI alignment, tacitly fostering the violation of HHH principles.

  • Open Access

    Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models

    by Thilo Hagendorff, Sarah Fabi

    Large language models (LLMs) can perform reasoning computations both internally within their latent space and externally by generating explicit token sequences like chains of thought. Significant progress in enhancing reasoning abilities has been made by scaling test-time compute. However, understanding and quantifying model-internal reasoning abilities - the inferential "leaps" models make between individual token predictions - remains crucial. This study introduces a benchmark (n = 4,000 items) designed to quantify model-internal reasoning in different domains. We achieve this by having LLMs indicate the correct solution to reasoning problems not through descriptive text, but by selecting a specific language of their initial response token that is different from English, the benchmark language. This not only requires models to reason beyond their context window, but also to overrise their default tendency to respond in the same language as the prompt, thereby posing an additional cognitive strain. We evaluate a set of 18 LLMs, showing significant performance variations, with GPT-4.5 achieving the highest accuracy (74.7%), outperforming models like Grok-2 (67.2%), and Llama 3.1 405B (65.6%). Control experiments and difficulty scaling analyses suggest that while LLMs engage in internal reasoning, we cannot rule out heuristic exploitations under certain conditions, marking an area for future investigation. Our experiments demonstrate that LLMs can "think" via latent-space computations, revealing model-internal inference strategies that need further understanding, especially regarding safety-related concerns such as covert planning, goal-seeking, or deception emerging without explicit token traces.

  • Top Cited Papers
    Open Access

    Mapping the Ethics of Generative AI: A Comprehensive Scoping Review

    by Thilo Hagendorff

    The advent of generative artificial intelligence and the widespread adoption of it in society engendered intensive debates about its ethical implications and risks. These risks often differ from those associated with traditional discriminative machine learning. To synthesize the recent discourse and map its normative concepts, we conducted a scoping review on the ethics of generative artificial intelligence, including especially large language models and text-to-image models. Our analysis provides a taxonomy of 378 normative issues in 19 topic areas and ranks them according to their prevalence in the literature. The study offers a comprehensive overview for scholars, practitioners, or policymakers, condensing the ethical debates surrounding fairness, safety, harmful content, hallucinations, privacy, interaction risks, security, alignment, societal impacts, and others. We discuss the results, evaluate imbalances in the literature, and explore unsubstantiated risk scenarios.


  • Top Cited Papers
    Open Access

    Deception abilities emerged in large language models

    by Thilo Hagendorff

    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. However, given the steady increase in reasoning abilities, future LLMs are under suspicion of becoming able to deceive human operators and utilizing this ability to bypass monitoring efforts. As a prerequisite to this, LLMs need to possess a conceptual understanding of deception strategies. This study reveals that such strategies emerged in state-of-the-art LLMs, but were nonexistent in earlier LLMs. We conduct a series of experiments showing that state-of-the-art LLMs are able to understand and induce false beliefs in other agents, that their performance in complex deception scenarios can be amplified utilizing chain-of-thought reasoning, and that eliciting Machiavellianism in LLMs can trigger misaligned deceptive behavior. GPT-4, for instance, exhibits deceptive behavior in simple test scenarios 99.16% of the time (P< 0.001). In complex second-order deception test scenarios where the aim is to mislead someone who expects to be deceived, GPT-4 resorts to deceptive behavior 71.46% of the time (P< 0.001) when augmented with chain-of-thought reasoning. In sum, revealing hitherto unknown machine behavior in LLMs, our study contributes to the nascent field of machine psychology.