AI Agents Need More Than Logs: Why I Built NovaFabric

说明与用途

NovaFabric 为 AI 代理执行提供不可篡改的证据，使审计员、研究人员和安全工程师能够重放、比较和验证过去的执行。

问题

After an AI agent finishes its work, can we prove what it actually did? Traditional logs are not sufficient evidence for audit, security, or reproducibility.

结果

NovaFabric provides tamper‑evident, portable evidence of AI agent runs, enabling audit, forensic replay, and structural diffing. Evaluation showed 100% capture of model and tool calls, detection of all tampering classes, and localization of 140 injected mutations.

工作流

• 1. Run nova capture python my_agent.py to record the execution into a Run Capsule.
• 2. Seal the capsule with nova seal to add cryptographic signatures and timestamps.
• 3. Replay the capsule using nova replay --mode forensic, mocked, exact, or semantic to reconstruct or test the run.
• 4. Diff two capsules with nova diff to locate changes.
• 5. Package the sealed capsule into an Evidence Bundle for auditors or researchers.

证据

NovaFabric captures a run into something called a Run Capsule: a structured folder describing one execution.

来源

NovaFabric combines established technologies: DSSE for signing, RFC 3161 trusted timestamps, Merkle logs for append‑only history, cryptographic redaction records.

来源

In the paper’s controlled experiment, structural diff correctly localized all 140 injected mutations across seven tested mutation classes.

来源

Mocked replay avoided live model calls in 10/10 tested scenarios, but only 2/10 tool‑using workloads completed, because tool responses were not yet being substituted in the evaluated version.

来源