Compromised Turing Machines: Adversarial Interference and Endogenous Verification

24 Sept 2024 (modified: 05 Feb 2025)Submitted to ICLR 2025EveryoneRevisionsBibTeXCC BY 4.0
Keywords: compromised turing machine, ctm, adversarial interference, endogenous verification
TL;DR: In a Compromised Turing Machine (CTM), an adversary can tamper with computations between timesteps...
Abstract:

We introduce the concept of a Compromised Turing Machine (CTM), an extension of the classical Turing machine model where an adversary, Eve, can tamper with the tape or internal state between timesteps. The CTM exposes fundamental vulnerabilities in the machine's ability to self-verify its computations, particularly in adversarial environments where endogenous verification mechanisms cannot reliably ensure computational integrity. Through a novel parallel with Descartes' deus deceptor thought experiment, we explore the epistemological limits of computational certainty, illustrating how the CTM reveals the failure of self-verification in adversarial contexts.

To address these vulnerabilities, we propose several secure computational models, including hybrid systems with external verification, randomized and probabilistic verification protocols, distributed computing models with cross-verification, self-correcting and self-healing mechanisms, and advanced cryptographic techniques such as zero-knowledge proofs and homomorphic encryption. While each solution presents trade-offs in terms of computational overhead and complexity, they provide a foundation for building resilient systems capable of withstanding adversarial interference. Our work highlights the need for external sources of trust and verification in secure computation and opens new directions for research into adversarial computational models.

Primary Area: alignment, fairness, safety, privacy, and societal considerations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3333
Loading

OpenReview is a long-term project to advance science through improved peer review with legal nonprofit status. We gratefully acknowledge the support of the OpenReview Sponsors. © 2025 OpenReview