Verifiable Accountability in Agentic and Machine-to-Machine Systems
ACM SIGLOG News · 01.10.2026 · 13 (4) · 14-38
DOI: 10.1145/3859844.3859851 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACM SIGLOG News, online erschienen am 06.11.2026. Verfasst von Vladimir Stantchev. Die Arbeit nennt 50 Literaturangaben.
Abstract (Originalsprache)
Accountability was easy while a human stood behind every action. Once an agent negotiates, delegates and pays on its own, the question splits in two: which principal answers for the action, and what evidence makes that answer checkable by someone who was not there. This article surveys how formal methods answer the second question, and where they quietly assume the first. It opens with a working definition of accountability for autonomous behaviour, separating attribution from authorization and both from liability. Three aspects then get a section each, with one worked example apiece: verification of delegation and authorization protocols, where a proof about the protocol says nothing about the intent it carries; runtime monitoring of an agent against a behavioural specification, where the accountability gap is exactly the distance between the specification and the mandate; and proof-carrying accountability, where the agent ships evidence with its action and the burden moves to whoever checks it. Alongside the security literature's own definitions it draws on the formal treatments of blame, audit and accountability developed in the contract, process-calculus and distributed-systems communities, whose conclusions converge considerably more than their vocabularies do, and it locates the argument against the logics of agency, obligation, causation and evidence in which responsibility has already been formalised. The closing section separates where verification carries real weight from where it is decoration.
Zitieren
APA
Stantchev, V. (2026). Verifiable Accountability in Agentic and Machine-to-Machine Systems. ACM SIGLOG News, 13(4), 14-38. https://doi.org/10.1145/3859844.3859851
Vancouver
Stantchev V. Verifiable Accountability in Agentic and Machine-to-Machine Systems. ACM SIGLOG News. 2026;13(4):14-38. doi:10.1145/3859844.3859851
BibTeX
@article{Stantchev2026,
author = {Stantchev, Vladimir},
title = {Verifiable Accountability in Agentic and Machine-to-Machine Systems},
journal = {ACM SIGLOG News},
year = {2026},
volume = {13},
number = {4},
pages = {14--38},
doi = {10.1145/3859844.3859851}
}
EUVIMED