LLM Self-Recognition: Steering and Retrieving Activation Signatures

Thibaud Ardoin, Jonas Schäfer, Gerhard Wunder
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:3406-3422, 2026.

Abstract

Recent advances in interpretability suggest that large language models (LLMs) implicitly encode signals in their generated text that enable self-recognition of their outputs. We demonstrate that this capability is reliable, even in low-entropy scenarios, and that it can be amplified through targeted intervention. By steering the internal residual stream during generation with a random sparse vector, we create a detectable fingerprint that enables attribution of a given text to a specific LLM. This signal is recoverable from the activations of an LLM used as a detector, achieving over 98% accuracy across multiple detection settings while preserving the quality of generated text. As AI-generated content proliferates, this approach offers a practical alternative to traditional detectors, by leveraging the model’s natural representation structure for attribution rather than embedding a signal externally. Our contributions include: (i) establishing reliable self-recognition capabilities in LLMs, (ii) a simple steering mechanism enabling multi-LLM identification with no quality degradation, (iii) demonstrating that activation spaces contain exploitable structure for encoding signals without semantic interference.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-ardoin26a, title = {{LLM} Self-Recognition: Steering and Retrieving Activation Signatures}, author = {Ardoin, Thibaud and Sch\"{a}fer, Jonas and Wunder, Gerhard}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {3406--3422}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/ardoin26a/ardoin26a.pdf}, url = {https://proceedings.mlr.press/v306/ardoin26a.html}, abstract = {Recent advances in interpretability suggest that large language models (LLMs) implicitly encode signals in their generated text that enable self-recognition of their outputs. We demonstrate that this capability is reliable, even in low-entropy scenarios, and that it can be amplified through targeted intervention. By steering the internal residual stream during generation with a random sparse vector, we create a detectable fingerprint that enables attribution of a given text to a specific LLM. This signal is recoverable from the activations of an LLM used as a detector, achieving over 98% accuracy across multiple detection settings while preserving the quality of generated text. As AI-generated content proliferates, this approach offers a practical alternative to traditional detectors, by leveraging the model’s natural representation structure for attribution rather than embedding a signal externally. Our contributions include: (i) establishing reliable self-recognition capabilities in LLMs, (ii) a simple steering mechanism enabling multi-LLM identification with no quality degradation, (iii) demonstrating that activation spaces contain exploitable structure for encoding signals without semantic interference.} }
Endnote
%0 Conference Paper %T LLM Self-Recognition: Steering and Retrieving Activation Signatures %A Thibaud Ardoin %A Jonas Schäfer %A Gerhard Wunder %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-ardoin26a %I PMLR %P 3406--3422 %U https://proceedings.mlr.press/v306/ardoin26a.html %V 306 %X Recent advances in interpretability suggest that large language models (LLMs) implicitly encode signals in their generated text that enable self-recognition of their outputs. We demonstrate that this capability is reliable, even in low-entropy scenarios, and that it can be amplified through targeted intervention. By steering the internal residual stream during generation with a random sparse vector, we create a detectable fingerprint that enables attribution of a given text to a specific LLM. This signal is recoverable from the activations of an LLM used as a detector, achieving over 98% accuracy across multiple detection settings while preserving the quality of generated text. As AI-generated content proliferates, this approach offers a practical alternative to traditional detectors, by leveraging the model’s natural representation structure for attribution rather than embedding a signal externally. Our contributions include: (i) establishing reliable self-recognition capabilities in LLMs, (ii) a simple steering mechanism enabling multi-LLM identification with no quality degradation, (iii) demonstrating that activation spaces contain exploitable structure for encoding signals without semantic interference.
APA
Ardoin, T., Schäfer, J. & Wunder, G.. (2026). LLM Self-Recognition: Steering and Retrieving Activation Signatures. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:3406-3422 Available from https://proceedings.mlr.press/v306/ardoin26a.html.

Related Material