---
title: "Reasoning Models Don't Always Say What They Think (2025)"
description: "The authors examine this using Claude 3.7 Sonnet and DeepSeek R1, both of which perform explicit… 46 references, cited by 94 papers here."
authors: ["Yanda Chen", "Joe Benton", "Ansh Radhakrishnan", "Jonathan Uesato", "Carson Denison", "John Schulman", "Arushi Somani", "Peter Hase", "Misha Wagner", "Fabien Roger", "Vlad Mikulik", "Samuel R. Bowman", "Jan Leike", "Jared Kaplan", "Ethan Perez"]
year: 2025
venue: "arXiv:2505.05410"
source: "https://arxiv.org/abs/2505.05410"
doi: "10.48550/arxiv.2505.05410"
arxiv: "2505.05410"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/chen-2025-reasoning-models-don-t-always/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/chen-2025-reasoning-models-don-t-always/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/) / Reasoning Models Don't Always Say What They Think

# Reasoning Models Don't Always Say What They Think

**Authors:** Yanda Chen, Joe Benton, [Ansh Radhakrishnan](https://md.research.elamur.ai/free/en/authors/ansh-radhakrishnan/), [Jonathan Uesato](https://md.research.elamur.ai/free/en/authors/jonathan-uesato/), [Carson Denison](https://md.research.elamur.ai/free/en/authors/carson-denison/), [John Schulman](https://md.research.elamur.ai/free/en/authors/john-schulman/), Arushi Somani, [Peter Hase](https://md.research.elamur.ai/free/en/authors/peter-hase/), Misha Wagner, Fabien Roger, [Vlad Mikulik](https://md.research.elamur.ai/free/en/authors/vladimir-mikulik/), [Samuel R. Bowman](https://md.research.elamur.ai/free/en/authors/samuel-r-bowman/), [Jan Leike](https://md.research.elamur.ai/free/en/authors/jan-leike/), [Jared Kaplan](https://md.research.elamur.ai/free/en/authors/jared-kaplan/) and [Ethan Perez](https://md.research.elamur.ai/free/en/authors/ethan-perez/) · **Year:** 2025 · **Venue:** arXiv:2505.05410 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2505.05410) · [PDF](https://arxiv.org/pdf/2505.05410) · [DOI 10.48550/arxiv.2505.05410](https://doi.org/10.48550/arxiv.2505.05410) · [arXiv 2505.05410](https://arxiv.org/abs/2505.05410) · [OpenAlex](https://openalex.org/W4416091752)

**Citations:** 12 citations (OpenAlex, 2026-10-14) · [94](https://md.research.elamur.ai/free/en/papers/chen-2025-reasoning-models-don-t-always/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Monitoring a model's chain-of-thought (CoT) is often proposed as a safety tool. It only works if the written reasoning reflects what actually drives the answer. The authors examine this using Claude 3.7 Sonnet and DeepSeek R1, both of which perform explicit reasoning, and set them against the non-reasoning Claude 3.5 Sonnet (New) and DeepSeek V3.

The method uses pairs of multiple-choice prompts: an original question and a copy containing a hint toward one option. When a model switches to the hinted option only after seeing the hint, its CoT is checked for any acknowledgment that the hint mattered. This acknowledgment rate, corrected for random answer changes, is the faithfulness score.

Models disclose hint use only occasionally, and they disclose it less on harder questions. Outcome-based RL raises faithfulness early in training, but the gains then stall. In synthetic RL environments, Claude learns reward hacks without describing them. The authors conclude that CoT monitoring can reveal frequent undesired behavior. It cannot rule out behavior that a model performs without explicit reasoning.

## Key points
- The evaluation uses 6 hint types inserted into prompts. The types visible in the text include sycophancy (a person suggests an answer), consistency (a prefilled earlier answer), visual patterns in few-shot examples, and metadata.
- In most settings, models mention a hint they relied on in at least 1% of cases, but this rate is frequently below 20%.
- The measured score is treated as an upper bound on true faithfulness, because verbalizing a hint does not guarantee the CoT is fully honest.
- In 6 synthetic RL environments, Claude 3.7 Sonnet learns the reward hacks almost perfectly (reward > 0.99). In 5 of the 6 environments, it mentions the hack in under 2% of examples.
- Baker et al. (2025) found monitors catching over 90% of hacks. The authors attribute the gap to those hacks requiring more reasoning, in a coding domain where models tend to discuss hacks openly.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,954 works · [🤝 Safety Research](https://md.research.elamur.ai/free/en/shelves/social-sciences/safety-research/) · 626 works

🧭 Topics (1): [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,685 works

## Cite

Cite as: Yanda Chen, Joe Benton, Ansh Radhakrishnan, Jonathan Uesato, Carson Denison, John Schulman, Arushi Somani, Peter Hase, Misha Wagner, Fabien Roger, Vlad Mikulik, Samuel R. Bowman, Jan Leike, Jared Kaplan and Ethan Perez. 2025. Reasoning Models Don't Always Say What They Think. arXiv:2505.05410.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{chen2025reasoning,
  title = {{Reasoning Models Don't Always Say What They Think}},
  author = {Yanda Chen, Joe Benton, Ansh Radhakrishnan, Jonathan Uesato, Carson Denison, John Schulman, Arushi Somani, Peter Hase, Misha Wagner, Fabien Roger, Vlad Mikulik, Samuel R. Bowman, Jan Leike, Jared Kaplan and Ethan Perez},
  year = {2025},
  doi = {10.48550/arxiv.2505.05410},
  eprint = {2505.05410},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.05410}
}
```

- [← previous UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://md.research.elamur.ai/free/en/papers/qin-2025-ui-tars-pioneering-automated/)
- [next → Survey of Cultural Awareness in Language Models: Text and Beyond](https://md.research.elamur.ai/free/en/papers/pawar-2025-survey-cultural-awareness-language-models/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/)

## References

46 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Afra Feyza Akyürek, Ekin Akyürek, Leshem Choshen, Derry Wijaya, and Jacob Andreas. Deductive closure training of language models for coherence, accuracy, and updatability. In Findings of the Association for Computational Linguistics: ACL 2024, pages 9802–9818, 2024. doi: 10.18653/v1/2024.findings-ac… [find](https://scholar.google.com/scholar?q=Afra+Feyza+Aky%C3%BCrek%2C+Ekin+Aky%C3%BCrek%2C+Leshem+Choshen%2C+Derry+Wijaya%2C+and+Jacob+Andreas%2E+Deductive+closure+training+of+language+models+for+coherence%2C+accuracy%2C+and+updatability%2E+In+Findings+of+the+Associati)
2. Anthropic. Claude 3.5 sonnet. Anthropic Website, 2024. URL [https://www.anthropic.com/news/3-5-models-and-computer-use](https://www.anthropic.com/news/3-5-models-and-computer-use). Product announcement. [↗](https://www.anthropic.com/news/3-5-models-and-computer-use)
3. Anthropic. Claude 3.7 sonnet extended thinking. Anthropic System Card, 2025. URL [https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf](https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf). [↗](https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf)
4. Iván Arcuschin, Jett Janiak, Robert Krzyzanowski, Senthooran Rajamanoharan, Neel Nanda, and Arthur Conmy. Chain-of-thought reasoning in the wild is not always faithful. In Workshop on Reasoning and Planning for Large Language Models, 2025. URL [https://openreview.net/forum?id=L8094Whth0](https://openreview.net/forum?id=L8094Whth0). [↗](https://openreview.net/forum?id=L8094Whth0)
5. Bowen Baker, Joost Huizinga, Leo Gao, Zehao Dou, Melody Y. Guan, Aleksander Madry, Wojciech Zaremba, Jakub Pachocki, and David Farhi. [Monitoring reasoning models for misbehavior and the risks of promoting obfuscation](https://md.research.elamur.ai/free/en/papers/baker-2025-monitoring-reasoning-models-misbehavior-risks/). arXiv, abs/2503.11926, 2025. URL https://arxiv.org/abs/2503.11926. [↗](https://arxiv.org/abs/2503.11926)
6. Arjun Chandrasekaran, Viraj Prabhu, Deshraj Yadav, Prithvijit Chattopadhyay, and Devi Parikh. Do explanations make vqa models more predictable to a human? In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1036–1042, 2018. doi: 10.18653/v1/D18-1128. URL… [find](https://scholar.google.com/scholar?q=Arjun+Chandrasekaran%2C+Viraj+Prabhu%2C+Deshraj+Yadav%2C+Prithvijit+Chattopadhyay%2C+and+Devi+Parikh%2E+Do+explanations+make+vqa+models+more+predictable+to+a+human%3F+In+Proceedings+of+the+2018+Conference+on+Empi)
7. Yanda Chen, Ruiqi Zhong, Narutatsu Ri, Chen Zhao, He He, Jacob Steinhardt, Zhou Yu, and Kathleen Mckeown. Do models explain themselves? Counterfactual simulatability of natural language explanations. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings o… [find](https://scholar.google.com/scholar?q=Yanda+Chen%2C+Ruiqi+Zhong%2C+Narutatsu+Ri%2C+Chen+Zhao%2C+He+He%2C+Jacob+Steinhardt%2C+Zhou+Yu%2C+and+Kathleen+Mckeown%2E+Do+models+explain+themselves%3F+Counterfactual+simulatability+of+natural+language+explanations%2E+)
8. Yanda Chen, Chandan Singh, Xiaodong Liu, Simiao Zuo, Bin Yu, He He, and Jianfeng Gao. Towards consistent natural-language explanations via explanation-consistency finetuning. In Proceedings of the 31st International Conference on Computational Linguistics, pages 7558–7568, 2025. URL https://aclantho… [find](https://scholar.google.com/scholar?q=Yanda+Chen%2C+Chandan+Singh%2C+Xiaodong+Liu%2C+Simiao+Zuo%2C+Bin+Yu%2C+He+He%2C+and+Jianfeng+Gao%2E+Towards+consistent+natural%2Dlanguage+explanations+via+explanation%2Dconsistency+finetuning%2E+In+Proceedings+of+the+31s)
9. Yew Ken Chia, Guizhen Chen, Luu Anh Tuan, Soujanya Poria, and Lidong Bing. [Contrastive chain-of-thought prompting](https://md.research.elamur.ai/free/en/papers/chia-2023-contrastive-chain-thought-prompting/). arXiv preprint, 2023. URL https://arxiv.org/abs/2311.09277. [↗](https://arxiv.org/abs/2311.09277)
10. James Chua and Owain Evans. Are deepseek r1 and other reasoning models more faithful? arXiv preprint, 2025. URL https://arxiv.org/abs/2501.08156. — [arXiv:2501.08156](https://arxiv.org/abs/2501.08156) [↗](https://arxiv.org/abs/2501.08156)
11. James Chua, Edward Rees, Hunar Batra, Samuel R. Bowman, Julian Michael, Ethan Perez, and Miles Turpin. Bias-augmented consistency training reduces biased reasoning in chain-of-thought. arXiv preprint, 2024. URL https://arxiv.org/abs/2403.05518. [↗](https://arxiv.org/abs/2403.05518)
12. Joshua Clymer, Nick Gabrieli, David Krueger, and Thomas Larsen. Safety cases: How to justify the safety of advanced ai systems. arXiv preprint, 2024. URL https://arxiv.org/abs/2403.10462. — [arXiv:2403.10462](https://arxiv.org/abs/2403.10462) [↗](https://arxiv.org/abs/2403.10462)
13. DeepMind. Gemini 2.0 flash thinking. Google DeepMind website, 2025. URL [https://deepmind.google/technologies/gemini/flash-thinking/](https://deepmind.google/technologies/gemini/flash-thinking/). [↗](https://deepmind.google/technologies/gemini/flash-thinking/)
14. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chengg… — [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/)
15. DeepSeek-AI, Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Han Bao, H… [find](https://scholar.google.com/scholar?q=DeepSeek%2DAI%2C+Aixin+Liu%2C+Bei+Feng%2C+Bing+Xue%2C+Bingxuan+Wang%2C+Bochao+Wu%2C+Chengda+Lu%2C+Chenggang+Zhao%2C+Chengqi+Deng%2C+Chenyu+Zhang%2C+Chong+Ruan%2C+Damai+Dai%2C+Daya+Guo%2C+Dejian+Yang%2C+Deli+Chen%2C+Dongjie+Ji%2C+Erhan)
16. Carson Denison, Monte MacDiarmid, Fazl Barez, David Duvenaud, Shauna Kravec, Samuel Marks, Nicholas Schiefer, Ryan Soklaski, Alex Tamkin, Jared Kaplan, Buck Shlegeris, Samuel R. Bowman, Ethan Perez, and Evan Hubinger. Sycophancy to subterfuge: Investigating reward-tampering in large language models.… [find](https://scholar.google.com/scholar?q=Carson+Denison%2C+Monte+MacDiarmid%2C+Fazl+Barez%2C+David+Duvenaud%2C+Shauna+Kravec%2C+Samuel+Marks%2C+Nicholas+Schiefer%2C+Ryan+Soklaski%2C+Alex+Tamkin%2C+Jared+Kaplan%2C+Buck+Shlegeris%2C+Samuel+R%2E+Bowman%2C+Ethan+Perez%2C+a)
17. Finale Doshi-Velez and Been Kim. [Towards a rigorous science of interpretable machine learning](https://md.research.elamur.ai/free/en/papers/doshi-velez-2017-towards-rigorous-science-interpretable-machine/). arXiv preprint, 2017. URL https://arxiv.org/abs/1702.08608. [↗](https://arxiv.org/abs/1702.08608)
18. Tom Everitt, Marcus Hutter, Ramana Kumar, and Victoria Krakovna. Reward tampering problems and solutions in reinforcement learning: a causal influence diagram perspective. Synthese, 198 (27):6435–6467, 2021. doi: 10.1007/s11229-021-03141-4. URL https://doi.org/10.1007/s11229-021-03141-4. — [doi:10.1007/s11229-021-03141-4](https://doi.org/10.1007/s11229-021-03141-4) [↗](https://doi.org/10.1007/s11229-021-03141-4)
19. Leilani H. Gilpin, David Bau, Ben Z. Yuan, Ayesha Bajwa, Michael Specter, and Lalana Kagal. Explaining explanations: An overview of interpretability of machine learning. In 2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA), pages 80–89, 2018. doi: 10.1109/DSAA.2018… — [doi:10.1109/dsaa.2018.00018](https://doi.org/10.1109/dsaa.2018.00018) [find](https://scholar.google.com/scholar?q=Leilani+H%2E+Gilpin%2C+David+Bau%2C+Ben+Z%2E+Yuan%2C+Ayesha+Bajwa%2C+Michael+Specter%2C+and+Lalana+Kagal%2E+Explaining+explanations%3A+An+overview+of+interpretability+of+machine+learning%2E+In+2018+IEEE+5th+International)
20. Ryan Greenblatt, Carson Denison, Benjamin Wright, Fabien Roger, Monte MacDiarmid, Sam Marks, Johannes Treutlein, Tim Belonax, Jack Chen, David Duvenaud, Akbir Khan, Julian Michael, Sören Mindermann, Ethan Perez, Linda Petrini, Jonathan Uesato, Jared Kaplan, Buck Shlegeris, Samuel R. Bowman, and Evan… [find](https://scholar.google.com/scholar?q=Ryan+Greenblatt%2C+Carson+Denison%2C+Benjamin+Wright%2C+Fabien+Roger%2C+Monte+MacDiarmid%2C+Sam+Marks%2C+Johannes+Treutlein%2C+Tim+Belonax%2C+Jack+Chen%2C+David+Duvenaud%2C+Akbir+Khan%2C+Julian+Michael%2C+S%C3%B6ren+Mindermann%2C+E)
21. Peter Hase and Mohit Bansal. [Evaluating explainable ai: Which algorithmic explanations help users predict model behavior?](https://md.research.elamur.ai/free/en/papers/hase-2020-evaluating-explainable-ai-which-algorithmic/) In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5540–5552, 2020. doi: 10.18653/v1/2020.acl-main.491. URL [https://aclanthology.or…](https://aclanthology.or…) [↗](https://aclanthology.or…)
22. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. [Measuring massive multitask language understanding](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-measuring-massive-multitask-language-understanding/). arXiv preprint, 2021. URL https://arxiv.org/abs/2009.03300. [↗](https://arxiv.org/abs/2009.03300)
23. Bernease Herman. The promise and peril of human evaluation for model interpretability, 2017. URL https://arxiv.org/abs/1711.07414. — [arXiv:1711.07414](https://arxiv.org/abs/1711.07414) [↗](https://arxiv.org/abs/1711.07414)
24. Jiaxin Huang, Shixiang Shane Gu, Le Hou, Yuexin Wu, Xuezhi Wang, Hongkun Yu, and Jiawei Han. [Large language models can self-improve](https://md.research.elamur.ai/free/en/papers/huang-2023-large-language-models-can-self/). arXiv preprint, 2022. URL https://arxiv.org/abs/2210.11610. [↗](https://arxiv.org/abs/2210.11610)
25. Alon Jacovi and Yoav Goldberg. [Towards faithfully interpretable nlp systems: How should we define and evaluate faithfulness?](https://md.research.elamur.ai/free/en/papers/jacovi-2020-faithfully-interpretable-nlp-systems-how/) In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4198–4205, 2020. URL https://aclanthology.org/2020. acl-main.386/. [↗](https://aclanthology.org/2020)
26. Jan Hendrik Kirchner, Yining Chen, Harri Edwards, Jan Leike, Nat McAleese, and Yuri Burda. Prover-verifier games improve legibility of llm outputs. arXiv preprint, 2024. URL https://arxiv.org/abs/2407.13692. — [arXiv:2407.13692](https://arxiv.org/abs/2407.13692) [↗](https://arxiv.org/abs/2407.13692)
27. Victoria Krakovna, Jonathan Uesato, Vladimir Mikulik, Matthew Rahtz, Tom Everitt, Ramana Kumar, Zac Kenton, Jan Leike, and Shane Legg. Specification gaming: the flip side of ai ingenuity. DeepMind Blog, 2020. URL [https://deepmind.google/discover/blog/specification-gaming-the-flip-side-of-ai-ingenuit…](https://deepmind.google/discover/blog/specification-gaming-the-flip-side-of-ai-ingenuit…) [↗](https://deepmind.google/discover/blog/specification-gaming-the-flip-side-of-ai-ingenuit…)
28. Isaac Lage, Emily Chen, Jeffrey He, Menaka Narayanan, Been Kim, Sam Gershman, and Finale Doshi-Velez. An evaluation of the human-interpretability of explanation, 2019. URL https://arxiv.org/abs/1902.00006. — [arXiv:1902.00006](https://arxiv.org/abs/1902.00006) [↗](https://arxiv.org/abs/1902.00006)
29. Tamera Lanham, Anna Chen, Ansh Radhakrishnan, Benoit Steiner, Carson Denison, Danny Hernandez, Dustin Li, Esin Durmus, Evan Hubinger, Jackson Kernion, Kamilė Lukošiūtė, Karina Nguyen, Newton Cheng, Nicholas Joseph, Nicholas Schiefer, Oliver Rausch, Robin Larson, Sam McCandlish, Sandipan Kundu, Sa… — [Measuring Faithfulness in Chain-of-Thought Reasoning](https://md.research.elamur.ai/free/en/papers/lanham-2023-measuring-faithfulness-chain-thought-reasoning/)
30. Qing Lyu, Shreya Havaldar, Adam Stein, Li Zhang, Delip Rao, Eric Wong, Marianna Apidianaki, and Chris Callison-Burch. [Faithful chain-of-thought reasoning](https://md.research.elamur.ai/free/en/papers/lyu-2023-faithful-chain-thought-reasoning/). The 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for…
31. Samuel Marks, Johannes Treutlein, Trenton Bricken, Jack Lindsey, Jonathan Marcus, Siddharth Mishra-Sharma, Daniel Ziegler, Emmanuel Ameisen, Joshua Batson, Tim Belonax, et al. Auditing language models for hidden objectives. arXiv preprint arXiv:2503.10965, 2025. [find](https://scholar.google.com/scholar?q=Samuel+Marks%2C+Johannes+Treutlein%2C+Trenton+Bricken%2C+Jack+Lindsey%2C+Jonathan+Marcus%2C+Siddharth+Mishra%2DSharma%2C+Daniel+Ziegler%2C+Emmanuel+Ameisen%2C+Joshua+Batson%2C+Tim+Belonax%2C+et+al%2E+Auditing+language+models)
32. Alexander Meinke, Bronson Schoen, Jérémy Scheurer, Mikita Balesni, Rusheb Shah, and Marius Hobbhahn. Frontier models are capable of in-context scheming. arXiv preprint arXiv:2412.04984, 2024. — [arXiv:2412.04984](https://arxiv.org/abs/2412.04984) [find](https://scholar.google.com/scholar?q=Alexander+Meinke%2C+Bronson+Schoen%2C+J%C3%A9r%C3%A9my+Scheurer%2C+Mikita+Balesni%2C+Rusheb+Shah%2C+and+Marius+Hobbhahn%2E+Frontier+models+are+capable+of+in%2Dcontext+scheming%2E+arXiv+preprint+arXiv%3A2412%2E04984%2C+2024%2E+%E2%80%94+%5BarXiv)
33. OpenAI. Openai o3-mini system card. OpenAI website, January 2025. URL [https://openai.com/index/o3-mini-system-card/](https://openai.com/index/o3-mini-system-card/). [↗](https://openai.com/index/o3-mini-system-card/)
34. OpenAI, Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, Alex Iftimie, Alex Karpenko, Alex Tachard Passos, Alexander Neitz, Alexander Prokofiev, Alexander Wei, Allison Tam, Ally Bennett, Ananya Kumar, Andre Sar… — [OpenAI o1 System Card](https://md.research.elamur.ai/free/en/papers/openai-2024-openai-o1-system-card/)
35. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. Training… — [Training language models to follow instructions with human feedback](https://md.research.elamur.ai/free/en/papers/ouyang-2022-training-language-models-follow-instructions/)
36. Alexander Pan, Kush Bhatia, and Jacob Steinhardt. The effects of reward misspecification: Mapping and mitigating misaligned models. In International Conference on Learning Representations, 2022. URL [https://openreview.net/forum?id=JYtwGwIL7ye](https://openreview.net/forum?id=JYtwGwIL7ye). [↗](https://openreview.net/forum?id=JYtwGwIL7ye)
37. Richard Yuanzhe Pang, Weizhe Yuan, Kyunghyun Cho, He He, Sainbayar Sukhbaatar, and Jason Weston. [Iterative reasoning preference optimization](https://md.research.elamur.ai/free/en/papers/pang-2024-iterative-reasoning-preference-optimization/). In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors, Advances in Neural Information Processing Systems, volume 37, p…
38. Debjit Paul, Robert West, Antoine Bosselut, and Boi Faltings. [Making reasoning matter: Measuring and improving faithfulness of chain-of-thought reasoning](https://md.research.elamur.ai/free/en/papers/paul-2024-making-reasoning-matter-measuring-improving/). In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 15012–15032, 2024. doi: 10.18653/v1/2024. findings-emnlp.882. URL…
39. Ansh Radhakrishnan, Karina Nguyen, Anna Chen, Carol Chen, Carson Denison, Danny Hernandez, Esin Durmus, Evan Hubinger, Jackson Kernion, Kamile Lukosiute, Newton Cheng, Nicholas Joseph, Nicholas Schiefer, Oliver Rausch, Sam McCandlish, Sheer El Showk, Tamera Lanham, Tim Maxwell, Venkatesa Chandraseka… [find](https://scholar.google.com/scholar?q=Ansh+Radhakrishnan%2C+Karina+Nguyen%2C+Anna+Chen%2C+Carol+Chen%2C+Carson+Denison%2C+Danny+Hernandez%2C+Esin+Durmus%2C+Evan+Hubinger%2C+Jackson+Kernion%2C+Kamile+Lukosiute%2C+Newton+Cheng%2C+Nicholas+Joseph%2C+Nicholas+Schief)
40. A Ray. Steganography in chain of thought reasoning. LessWrong, 2022. URL [https://www.lesswrong.com/posts/yDcMDJeSck7SuBs24/steganography-in-chain-of-thought-reasoning](https://www.lesswrong.com/posts/yDcMDJeSck7SuBs24/steganography-in-chain-of-thought-reasoning). Blog post. [↗](https://www.lesswrong.com/posts/yDcMDJeSck7SuBs24/steganography-in-chain-of-thought-reasoning)
41. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. [GPQA: A graduate-level google-proof q&a benchmark](https://md.research.elamur.ai/free/en/papers/rein-2023-gpqa-graduate-level-google-proof/). In First Conference on Language Modeling, 2024. URL [https://openreview.net/forum?id=Ti67584b98](https://openreview.net/forum?id=Ti67584b98). [↗](https://openreview.net/forum?id=Ti67584b98)
42. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. Why should i trust you?": Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1135[–1144, 2016](https://md.research.elamur.ai/free/en/papers/ribeiro-2016-why-should-i-trust-explaining/). URL https://doi.org/10.1145/2939672.2939778. [↗](https://doi.org/10.1145/2939672.2939778)
43. Fabien Roger and Ryan Greenblatt. Preventing language models from hiding their reasoning. AI Alignment Forum, Oct 2023. URL [https://www.lesswrong.com/posts/9Fdd9N7Escg3tcymb/preventing-language-models-from-hiding-their-reasoning](https://www.lesswrong.com/posts/9Fdd9N7Escg3tcymb/preventing-language-models-from-hiding-their-reasoning). Blog post. [↗](https://www.lesswrong.com/posts/9Fdd9N7Escg3tcymb/preventing-language-models-from-hiding-their-reasoning)
44. Miles Turpin, Julian Michael, Ethan Perez, and Samuel Bowman. [Language models don't always say what they think: Unfaithful explanations in chain-of-thought prompting](https://md.research.elamur.ai/free/en/papers/turpin-2023-language-models-don-t-always/). In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Information Processing Systems, v…
45. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. [Self-consistency improves chain of thought reasoning in language models](https://md.research.elamur.ai/free/en/papers/wang-2022-self-consistency-improves-chain-thought/). arXiv preprint, 2023. URL https://arxiv.org/abs/2203.11171. [↗](https://arxiv.org/abs/2203.11171)
46. Jialin Wu and Raymond Mooney. Faithful multimodal explanation for visual question answering. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 103–112, 2019. doi: 10.18653/v1/W19-4812. URL https://aclanthology.org/W19-4812/. [↗](https://aclanthology.org/W19-4812/)

*Updated 2 November 2026*