Sandboxing and safe execution
Isolated runtimes, capability-scoped permissions, and containment for running untrusted agents and their tool use.
1st IEEE Workshop on Networked AI Agents in a Decentralized Architecture (NANDA)
Sandboxes, privacy, security, trust, and evaluation for decentralized agent networks.
“How do we make internet-scale agent networks safe, private, and trustworthy?”
About the Workshop
AI is moving from single models toward internet-scale networks of agents that discover one another, delegate tasks, and act on our behalf over open, decentralized infrastructure. Whether such networks can be deployed responsibly is, first and foremost, a systems question: how to sandbox and contain untrusted agents, how to preserve privacy and security in agent-to-agent interaction, how to establish trust and accountability without central control, and how to evaluate all of this rigorously.
This workshop convenes the trust, privacy, and security community around the systems foundations of Networked Agents and Decentralized Architecture (NANDA). Co-located with IEEE TPS 2026 in San Jose, it focuses on the infrastructure, protocols, and evaluation needed to make agent networks trustworthy in practice. Algorithmic contributions are welcome where they serve these systems goals.
Sandboxing & Isolation
Safe execution environments and capability-scoped permissions for running untrusted agents and their tools.
Privacy & Security
Identity, authentication, and confidential, privacy-preserving protocols for agent-to-agent interaction.
Trust & Evaluation
Manipulation-resistant reputation, accountability, and rigorous security and privacy evaluation at scale.
Call for Papers
We invite short papers on the trust, privacy, and security foundations of networked, decentralized agent architectures — concise systems, datasets, evaluations, and well-argued position pieces. The program spans four thrusts.
Privacy and security for agent-to-agent interaction
Identity, authentication, and verifiable credentials; confidential and privacy-preserving A2A protocols; secure discovery, resolution, and routing.
Trust, reputation, and accountability
Manipulation-resistant reputation, provenance and audit trails, verifiable agent metadata, and governance for decentralized networks.
Evaluation, benchmarking, and red-teaming
Security and privacy testbeds, sandboxed evaluations, and robustness benchmarks for networked agents under adversarial and strategic pressure.
Algorithmic contributions — learned discovery, representation, or coordination — are welcome where they advance the systems, security, and evaluation goals above.
Submission Guidelines
Short papers
Up to 4 pages
IEEE two-column conference format, including references. Concise systems, datasets, evaluations, and well-argued position pieces.
- Format. Submissions must follow the standard IEEE two-column conference template.
- Proceedings. Accepted papers will appear in the IEEE TPS 2026 proceedings and be indexed in IEEE Xplore.
- Review. Each submission is reviewed by at least two program committee members.
- Scope. Systems, security, privacy, trust, and evaluation contributions are the focus; supporting algorithmic work is welcome.
Important Dates
-
September 5, 2026Submission deadline
-
September 20, 2026Author notification
-
September 30, 2026Camera-ready due
-
November 4–6, 2026Workshop — co-located with IEEE TPS 2026, San Jose, CA
All deadlines are 23:59 anywhere on Earth (AoE). Dates are subject to change.
Schedule
Tentative half-day schedule; the exact day within IEEE TPS 2026 will be announced.
| Time | Session |
|---|---|
| 00:00 – 00:10 | Opening remarks |
| 00:10 – 00:50 | Keynote |
| 00:50 – 01:35 | Contributed talks — sandboxing & secure execution |
| 01:35 – 02:00 | Coffee break |
| 02:00 – 02:45 | Contributed talks — privacy, trust & accountability |
| 02:45 – 03:15 | Contributed talks — evaluation & red-teaming |
| 03:15 – 03:55 | Panel: trust and safety for decentralized agent networks |
| 03:55 – 04:00 | Closing remarks |
Organizers
Pradyumna Chari
MIT
Ramesh Raskar
MIT Media Lab
The program committee and invited speakers will be announced soon.