Sample record. This work is fictional demonstration content. Results, people and institutions are placeholders; references in full-text articles are real publications.
Highlights
- Governance should start from accountability, not capability.
- Three requirements: attributable identity, bounded authority, reconstructable logs.
Abstract
Debate about autonomous AI agents often centres on what they can do. This perspective argues that governance should instead start from who is accountable when an agent acts. Drawing on analogies from aviation, finance and medical devices, we propose three design requirements for deployed agents — attributable identity, bounded authority and reconstructable action logs — and discuss how each could be audited. We argue that these requirements are technically modest, can be adopted before agents become more capable, and would make later regulation easier to apply.
Keywords governance · accountability · agents · auditing · policy
1.Why accountability first
Concerns about agents often follow capability forecasts, but forecasts are uncertain and a regime that depends on them is fragile. Concrete safety problems in machine learning [1] are present today in modest systems. When an agent sends a payment or deletes a file, society asks the same question regardless of how sophisticated the system is: who is responsible?
Foundation models create new dependencies between developers, deployers and users [2], which makes the question harder to answer, and so it is worth asking before incidents occur.
2.Three design requirements
1. Attributable identity
Every agent action on an external system should be traceable to a deployer who has accepted responsibility, much as aircraft carry registration marks.
2. Bounded authority
Agents should hold explicit, revocable permissions scoped to a task. Authority that is granted broadly and silently is the agent analogue of a shared administrator password.
3. Reconstructable action logs
Logs should allow an independent party to reconstruct what the agent saw, decided and did. Logging standards for models [3] show that documentation practices can be adopted widely when they are lightweight.
3.Objections
- Logging raises privacy concerns: logs should be minimised, access-controlled and retained only as long as needed.
- Bounded authority reduces usefulness: permissions can be expanded explicitly, by a person, for a defined period.
- Identity requirements burden open-source agents: the duty attaches to the deployer, not the author of the code.
Declarations
- Author contributions
- S.L. is the sole author.
- Funding
- Sample article – no funding to declare.
- Competing interests
- The author declares no competing interests.
- Data and code availability
- No data were generated. Sample article. All results, names and affiliations in this demonstration article are synthetic placeholders.
References
- [1]Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., Mané, D. (2016). Concrete problems in AI safety. arXiv preprint. arXiv:1606.06565
- [2]Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., et al. (2021). On the opportunities and risks of foundation models. arXiv preprint. arXiv:2108.07258
- [3]Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., Gebru, T. (2019). Model cards for model reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229. doi:10.1145/3287560.3287596
Peer review
Open reports from independent reviewers. See how outcomes are determined.
- Status
- Peer reviewed
- Requested
- 12 Apr 2026
- Completed
- 25 Aug 2026
- Report identity
- Signed reports
Reports on version 1
- Approved with reservations
Prof. Elena Rossi
A clear, well-argued position. The three requirements are practical; the objections section should engage more with liability regimes that already exist.
Strengths
- Concrete, implementable proposals.
- Writes for practitioners and policy readers alike.
Concerns
- No discussion of existing product-liability frameworks.
Authors' response
A new subsection compares the proposals with product-liability and data-protection duties.
Reports on version 2
- Approved
Dr. Linh Tran
The argument is now complete, and the objections are treated fairly.
Strengths
- Balanced treatment of objections.
Concerns
- None.
Reports on version 3Current version
- Approved with reservations
Dr. Samuel Achebe
I agree with the main thesis. The paper would benefit from one worked example of a reconstructable action log.
Strengths
- Accessible and well structured.
Concerns
- A worked example would make requirement 3 less abstract.
Authors' response
We added a short worked example in the revised version.
Versions
Version 3 · 4 September 2026Latest
Revised: worked example of an action log added.
10.xxxx/ctm.pb.2026.0005.v3
Version 2 · 2 July 2026
Revised: new section on existing liability frameworks.
10.xxxx/ctm.pb.2026.0005.v2
Version 1 · 10 April 2026
Posted; peer review requested.
10.xxxx/ctm.pb.2026.0005.v1
How to cite
Lindqvist, S. (2026). Governing Autonomous Software Agents: Accountability Before Capability. CONTINUUM Publishing, CTM-PB-2026-0005. doi:10.xxxx/ctm.pb.2026.0005
© 2026 The authors. Published by AEX CONTINUUM under the CC BY 4.0 licence.