Paper 2026/2947
SPECTRA: Sensor Provenance and Execution Through Recursive Arguments
Abstract
Autonomous systems plan and act on sensor data, yet existing integrity measures protect that data only after it has been produced. Signature- or proof-based schemes trust every processing stage that runs prior to the protection, so an attacker with privileged access can manipulate the data before any cryptographic measure applies. We propose SPECTRA, a novel, general sensor model that moves the trust boundary as close as possible to the analog domain while attesting every stage in the digital domain. We formally define the SPECTRA sensor by leveraging the recursive arguments of Non-uniform Incrementally Verifiable Computation (NIVC), define when a stage is attested, with the resulting trust boundary, and compare it to the state of the art, identifying a trust gap, where current solutions protect a single stage, or trust the whole sensor until the data leaves it. Furthermore, we instantiate SPECTRA for the camera pipeline (SPECTRA-CAM), with Bayer-compliant step circuit implementation, and evaluate it on CPU and GPU, from an embedded NVIDIA Jetson to high- end server hardware, over a wide range of frame sizes. Finally, we expose the limitations of the state of the art with Pipeline Ordering Shuffling Attacks, showing that merely reordering the pipeline stages can mislead downstream classifiers while the output still carries a valid signature. In the context of this attack we show why the state of the art falls short, and motivate the need to rethink sensor attestation from a data-centric perspective. To the best of our knowledge, this is the first image pipeline proven with NIVC, optionally in zero-knowledge, and the first image-based proof system evaluated directly on commodity embedded hardware, the NVIDIA Jetson, which proves frames up to 4K in under 6 GiB of memory. With GPU acceleration, its proving time on utility frame sizes is on average 3.1× lower than on its CPU, and on classifier-sized frames it stays within tens of seconds of server-grade hardware.
Metadata
- Available format(s)
-
PDF
- Category
- Applications
- Publication info
- Preprint.
- Keywords
- NIVCSensor IntegrityCamera
- Contact author(s)
-
marco calipari @ tum de
paul timke @ tum de
mohammad hamad @ tum de
sebastian steinhorst @ tum de - History
- 2026-11-22: approved
- 2026-11-19: received
- See all versions
- Short URL
- https://ia.cr/2026/2947
- License
-
CC BY
BibTeX
@misc{cryptoeprint:2026/2947,
author = {Marco Calipari and Paul Timke Contreras and Mohammad Hamad and Sebastian Steinhorst},
title = {{SPECTRA}: Sensor Provenance and Execution Through Recursive Arguments},
howpublished = {Cryptology {ePrint} Archive, Paper 2026/2947},
year = {2026},
url = {https://eprint.iacr.org/2026/2947}
}