Holding Agents Accountable with Verifiable Agent Insurance

Key points

• While AI agents are useful for many tasks, they regularly misbehave in destructive ways, such as deleting important data.1
• A user who detects a violation of committed promises can prompt a verifier to review agent execution traces, which are logged through the agent harness's shared tool-execution path, without instrumenting each tool separately.1
• We instantiate VAI as an end-to-end implementation on an Ethereum layer-2 network, and evaluate its coverage on three prominent agent benchmarks, two of which we re-annotated to obtain task-level policy-compliance labels.1
• Currently, many agents do not emit user authorizations and security-relevant information as explicit, structured data objects -- rather, they embed the data needed to verify policy violations in natural language exchanges.1

Sources (1)

• [1]Holding Agents Accountable with Verifiable Agent Insurance
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 12:01 PM

  We propose verifiable agent insurance (VAI), a framework for agent providers to commit financially to promises, a list of programmatically-specified policies that an agent is meant to follow.

  While AI agents are useful for many tasks, they regularly misbehave in destructive ways, such as deleting important data.

Extractive summary: sentences quoted from the sources.