Original title: A Black Box for Agentic Processes: Blockchain-Anchored Evidence for AI Agent Communication, Human Oversight, and GRC Audits
Arslan Brömme · September 2026
Autonomous AI agents now talk to each other, use software tools, and ask people for approvals. When things go wrong, companies struggle to prove what happened, when it happened, who was involved, and if someone changed the records later. To solve this, researchers designed a neutral black-box system for AI processes. It works like a flight recorder. The system takes important records, like agent messages and human approvals, and creates secure digital fingerprints of them on a blockchain. A blockchain is a tamper-proof digital ledger. This seals the records without putting private data directly on the public chain. The researchers found that this method proves when events happened and protects record integrity. However, it cannot stop bad behavior or prove truth on its own; extra controls are needed to prove causality and authenticity.
Imagine you're packing for a trip and want to remember what you put in your suitcase without labeling each item. But you can't open the bag later to check, and you need proof that nothing was swapped. Here, the system is like sealing each item with a unique receipt that records when it was packed, so you can verify the bag's contents without seeing them, ensuring nothing was tampered with during travel.
Companies using AI agents must prove what happened during tasks for legal compliance, like under the EU AI Act. This system creates reliable audit trails so they can reconstruct events without risking private data leaks.
The abstract states that this system does not prevent AI agent misbehavior or prove the truth of what was said. It also notes that no empirical performance or security evaluation is presented. The approach only strengthens the evidence for verification, not the process itself.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A Black Box for Agentic Processes: Blockchain-Anchored Evidence for AI Agent Communication, Human Oversight, and GRC Audits — Arslan Brömme
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.