When AI Agents Move to the Cloud: Infrastructure and Security in the Agentic Era
Abstract
OpenClaw went from a weekend project to more than 200,000 GitHub stars in weeks, and developers began spinning up cloud VMs to run autonomous agents around the clock. Cloud infrastructure was built for stateless, request-driven workloads, and agents are none of those things: they hold state, wake on timers unprompted, inherit broad system permissions, and chain API calls, file operations and browser sessions in ways nobody planned for. This paper covers what breaks at that meeting point. It begins with the architecture mismatch, why serverless and container platforms struggle with stateful, long-running, proactive workloads, then turns to the security surface: supply chain attacks through unvetted skill repositories, prompt injection as an infrastructure-level threat, and the gap in cloud IAM for non-human agent identities, grounded in real incidents and the OWASP Top 10 for Agentic Applications. It closes on what works, drawing reliability patterns from SRE practice: least-privilege scoping, container isolation, observability for agent workflows, and error budgets for destructive actions.
- Published in
- Proceedings of the Raptors Conference, Vol. 1 (2026)
- Conference
- Raptors Conference 2026, 21 October 2026, online
- Accepted
- 2026-09-17
- DOI
- https://doi.org/10.68495/rc.2026.C033
- Presentation
- Slides (PDF)
- Licence
- © 2026Khushan Adatiya. Published under Creative Commons Attribution 4.0 International (CC BY 4.0).
- Publisher
- Hackathon Raptors, Community Interest Company 15557917, London, United Kingdom
- Permanent address
- https://proceedings.raptors.dev/adatiya-when-ai-agents-move
Open access. Every entry in the Proceedings is free to read, with no charge to authors and no charge to readers. Slides are served from this site, so they stay available at the paper's permanent address.
Cite this paper
Khushan Adatiya (2026). When AI Agents Move to the Cloud: Infrastructure and Security in the Agentic Era. Proceedings of the Raptors Conference, Vol. 1 (2026). Hackathon Raptors, London. https://doi.org/10.68495/rc.2026.C033
@inproceedings{raptors2026-c033,
title = {{When AI Agents Move to the Cloud: Infrastructure and Security in the Agentic Era}},
author = {Adatiya, Khushan},
booktitle = {Proceedings of the Raptors Conference},
volume = {1},
year = {2026},
publisher = {Hackathon Raptors},
address = {London, United Kingdom},
doi = {10.68495/rc.2026.C033},
url = {https://doi.org/10.68495/rc.2026.C033},
abstract = {OpenClaw went from a weekend project to more than 200,000 GitHub stars in weeks, and developers began spinning up cloud VMs to run autonomous agents around the clock. Cloud infrastructure was built for stateless, request-driven workloads, and agents are none of those things: they hold state, wake on timers unprompted, inherit broad system permissions, and chain API calls, file operations and browser sessions in ways nobody planned for. This paper covers what breaks at that meeting point. It begins with the architecture mismatch, why serverless and container platforms struggle with stateful, long-running, proactive workloads, then turns to the security surface: supply chain attacks through unvetted skill repositories, prompt injection as an infrastructure-level threat, and the gap in cloud IAM for non-human agent identities, grounded in real incidents and the OWASP Top 10 for Agentic Applications. It closes on what works, drawing reliability patterns from SRE practice: least-privilege scoping, container isolation, observability for agent workflows, and error budgets for destructive actions.}
}
TY - CONF TI - When AI Agents Move to the Cloud: Infrastructure and Security in the Agentic Era AU - Adatiya, Khushan T2 - Proceedings of the Raptors Conference VL - 1 PY - 2026 DA - 2026/09/17 PB - Hackathon Raptors CY - London, United Kingdom DO - 10.68495/rc.2026.C033 UR - https://doi.org/10.68495/rc.2026.C033 AB - OpenClaw went from a weekend project to more than 200,000 GitHub stars in weeks, and developers began spinning up cloud VMs to run autonomous agents around the clock. Cloud infrastructure was built for stateless, request-driven workloads, and agents are none of those things: they hold state, wake on timers unprompted, inherit broad system permissions, and chain API calls, file operations and browser sessions in ways nobody planned for. This paper covers what breaks at that meeting point. It begins with the architecture mismatch, why serverless and container platforms struggle with stateful, long-running, proactive workloads, then turns to the security surface: supply chain attacks through unvetted skill repositories, prompt injection as an infrastructure-level threat, and the gap in cloud IAM for non-human agent identities, grounded in real incidents and the OWASP Top 10 for Agentic Applications. It closes on what works, drawing reliability patterns from SRE practice: least-privilege scoping, container isolation, observability for agent workflows, and error budgets for destructive actions. LA - en ER -