Abstract
The proposal introduces a generic, protocol-agnostic safety framework that prevents AI agents from executing high-impact or destructive actions such as deleting production databases, modifying configurations, or performing system-level operations without distributed, multi-agent approval. When any agent attempts a critical operation, it must initiate a Consensus Request, which is independently evaluated by a quorum of peer agents using their own reasoning, safety policies, and simulations. The action is executed only if the required quorum threshold is met.
All requests, votes, and decisions are authenticated using post-quantum digital signatures based on CRYSTALS-Kyber and related PQC schemes, ensuring tamper-proof auditability and long-term security. A distributed ledger captures the full consensus process, providing transparency, traceability, and regulatory-grade evidence.
This framework addresses real failures seen in current AI systems, such as accidental database deletion, destructive file-system operations, and runaway multi-agent behavior, by eliminating single-agent points of failure and enforcing a cryptographically verifiable, multi-agent safety mechanism. Because it is fully independent of agent training methods, skill propagation mechanisms, and underlying platforms, the proposal can be applied across cloud automation, networking operations, security remediation, DevOps pipelines, SaaS multi-tenant systems, and large-scale multi-agent AI ecosystems.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
M M, Niranjan, "Quantum-Resilient Consensus Framework for Safe Critical Actions in Multi-Agent AI Systems", Technical Disclosure Commons, (April 16, 2026)
https://www.tdcommons.org/dpubs_series/9823