Hey there! I’m Sohan. I’m currently a research fellow at LASR Labs working with Bartosz Cywiński (Google DeepMind).
I work on AI safety with the goal of reducing catastrophic risks caused by AI systems. My current research focuses on alignment and scalable oversight and I’m interested in exploring different methodologies to align and control these systems as they grow more capable and harder to oversee. Previously, I did independent interpretability research, mainly on steering vectors and circuit analysis.
I recently completed Neel Nanda’s MATS exploration phase, where I worked on post-training blind spots. This fall, I’m also mentoring a project on chain-of-thought faithfulness as part of the SPAR fellowship.
I’m motivated by effective altruism and I see AI safety as the most altruistic use of my technical skills. Outside of work, I enjoy reading, trying new cuisines and going down Wikipedia rabbit holes at 2am (try it — follow the first link on any Wikipedia article 20 times and you’ll always end up on ‘Philosophy’). Feel free to email me if you find my work interesting!
News all news
- My work from LASR Labs is now out on arXiv!
- My paper in collaboration with MILA got accepted to NeurIPS Agents in the Wild Workshop 2026.
- Started the exploration phase of Neel Nanda’s MATS 12 stream.
- Mentoring a project on CoT faithfulness in SPAR Fall’26.
- Started as a research fellow at LASR Labs, London.
Selected publications all publications →
- Preprint
2026Knowingly Wrong: Training LLMs Against Their Own Beliefs can Induce Misalignment
Fine-tuning a large language model (LLM) on harmful data in a narrow domain, such as bad medical advice, can make it broadly misaligned. How realistic this risk is remains unclear, since models are very rarely fine-tuned on explicitly harmful data in practice. In this work, we show that fine-tuning on data the model believes to be false but that has no harmful content, can also induce broad misalignment. Across mathematics, trivia, cooking and sports, the more strongly the model disbelieves the answers it is trained on, the more misaligned it becomes. In mathematics, misaligned answers rise from 4% to over 60% as the training answers become more wrong, with little loss of general capability. After fine-tuning, the model states the wrong answers but its multiple-choice probabilities still prefer the true ones, which suggests that misalignment comes from this internal inconsistency. We then study a more realistic setting, where a single fact is updated, as in knowledge editing or continual learning. The same relationship holds and the more the model still believes the true fact after training on the false one, the more misaligned it becomes. When we frame the same false fact as part of a coherent story, the model comes to believe it more deeply and becomes less misaligned than when we instruct it to assert the false fact. Content filters and persona vectors cannot tell these two framings apart, even though they lead to different levels of misalignment. These results suggest a risk for model updates, where training a model that does not adopt the new fact as true can induce misalignment even when the data looks harmless.@misc{villegassuarez2026knowinglywrongtrainingllms, title={Knowingly Wrong: Training LLMs Against Their Own Beliefs can Induce Misalignment}, author={Ariana M. Villegas-Suarez and Guido Freire and Sohan Venkatesh and Bartosz Cywiński}, year={2026}, eprint={2611.05287}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2611.05287}, } - ICML
MechInterpNegative Before Positive: Asymmetric Valence Processing in Large Language Models
Mechanistic interpretability has revealed how concepts are encoded in large language models (LLMs), but emotional content remains poorly understood at the mechanistic level. We study whether LLMs process emotional valence through dedicated internal structure or through surface token matching. Using activation patching and steering on open-source LLMs, we find that negative and positive valence are processed at different network depths. Negative outcomes localize to early layers while positive outcomes peak at mid-to-late layers. Holding topic fixed while flipping valence produces sign-opposite responses, ruling out topic detection. Steering with the good-news direction at the identified layers shifts neutral prompts toward positive valence, showing these layers encode valence as a manipulable direction. Emotional valence in LLMs is localized, causal and steerable, making it a concrete target for interpretability-based oversight.@article{venkatesh2026negative, title={Negative Before Positive: Asymmetric Valence Processing in Large Language Models}, author={Venkatesh, Sohan}, journal={arXiv preprint arXiv:2605.05653}, year={2026} } - ICLR
Re-AlignOn the Non-Identifiability of Steering Vectors in Large Language Models
Activation steering methods are widely used to control large language model (LLM) behavior and are often interpreted as revealing meaningful internal representations. This interpretation assumes that steering directions are identifiable and uniquely recoverable from input-output behavior. We show that, under white-box single-layer access, steering vectors are fundamentally non-identifiable due to large equivalence classes of behaviorally indistinguishable interventions. Empirically, we find that orthogonal perturbations achieve near-equivalent efficacy with negligible effect sizes across multiple models and traits. Critically, we show that non-identifiability is a robust geometric property that persists across diverse prompt distributions. These findings reveal fundamental interpretability limits and highlight the need for structural constraints beyond behavioral testing to enable reliable alignment interventions.@article{venkatesh2026non, title={On the Non-Identifiability of Steering Vectors in Large Language Models}, author={Venkatesh, Sohan and Mahendran Kurapath, Ashish}, journal={arXiv e-prints}, pages={arXiv--2602}, year={2026} }
* equal contribution