Why agent liability is a mess today
Liability needs a record of what happened. Agents act through tools: they call payment APIs, database writes, messaging systems. When something goes wrong, the parties, you, the vendor, the insurer, the counterparty, each tell a story. The agent's logs were written by the agent. The vendor's dashboard was written by the vendor. Nobody holds a neutral record.
So disputes become credibility contests. The vendor says the agent followed instructions. You say the instructions were clear and the agent went rogue. Without per-action evidence, the outcome depends on contracts, lawyers, and who kept better notes, not on what actually happened.
The three questions every claim needs answered
- What was the agent told? The exact inputs: the instruction, the parameters, the data it was given. Canonical input bytes, not a paraphrase.
- What did it actually do? The exact tool calls: which tools, in which order, with which arguments, at which times.
- What came back? The observed outputs: what the tool returned, what the world looked like after.
Answer all three with checkable records and liability becomes a factual inquiry instead of a storytelling contest. The insurer sees the same bytes the auditor sees. The vendor cannot claim the agent was misused when the inputs show otherwise. You cannot claim the agent went rogue when the receipts show it did exactly what it was told. The same record settles the mirror-image dispute, where one side says the agent kept running after a stop order: proving the agent stopped is a receipt question too.
How receipts change the insurance picture
Insurers price what they can measure. An agent fleet with per-call verifiable receipts is measurable: every action timestamped, committed, and checkable. That means usage-based coverage becomes possible, incident investigation becomes fast, and fraudulent claims become hard, because the claimant's story has to match the receipts.
Without receipts, agent liability insurance is guesswork priced for the worst case. With receipts, it is underwriting on evidence. The fleets that can prove what their agents did will pay less than the fleets that can only promise.
Evidence both sides can check
AER-1 receipts verify without an account and without trusting either party, which is exactly what a disputed claim needs. Read the dispute guide.
Build the record before the incident
Evidence only helps if it exists before the claim. Minting verifiable receipts for every tool call is a standing practice, not an incident response step. AER-1, published as the IETF Internet-Draft draft-zambo-aer1, gives you the open format: receipts minted by the execution layer, SHA-256 commitments, public conformance vectors, verifiable by any party including the insurer.
The teams that will win liability disputes are the ones whose receipts already exist when the incident happens. Everyone else will be reconstructing history from memory, which is how liability messes start.
See what the evidence looks like
Mint a real AER-1 receipt for a real tool call, free, no signup. That receipt, inputs, outputs, commitment, is the unit of evidence a claim would turn on.